arxiv: 2605.03698 · v1 · submitted 2026-05-05 · 🧮 math.ST · stat.TH

Recognition: unknown

LAN property for the parameter of the jump rate in mean field interacting systems of neurons

Aline Duarte, Aur\'elien Velleret, Dasha Loukianova

Pith reviewed 2026-05-07 12:32 UTC · model grok-4.3

classification 🧮 math.ST stat.TH

keywords local asymptotic normalitymean-field limitinteracting particle systemsneuron spiking modelsjump rate estimationmaximum likelihood estimatorasymptotic minimax optimalityparameter estimation

0 comments

The pith

The maximum likelihood estimator for the spiking rate parameter in mean-field neuron systems is asymptotically efficient as the number of neurons tends to infinity.

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

This paper establishes that a multidimensional parameter in the spiking rate of neurons can be estimated efficiently from observations in the mean-field limit of many interacting neurons. The authors prove the local asymptotic normality property for the likelihood, which implies that the maximum likelihood estimator achieves the lowest possible asymptotic variance. For models that include resets of neuron states after each spike, they derive consistency, asymptotic normality, and local asymptotic minimax optimality using a separate classical argument. A sympathetic reader would care because these results supply a rigorous basis for reliable parameter inference from neural activity data when the system size is large.

Core claim

In a system of N neurons interacting in a mean-field regime as N tends to infinity, with states observed over a fixed time horizon, the local asymptotic normality property holds for the parameter in the jump rate. This property establishes the asymptotic efficiency of the maximum likelihood estimator. In the general setting that includes neuron resets at spike times, the estimator is consistent, asymptotically normal, and locally asymptotically minimax optimal.

What carries the argument

The local asymptotic normality property of the log-likelihood in the mean-field limit of the interacting particle system, which transfers classical efficiency and optimality results to the jump rate parameter.

Load-bearing premise

The mean-field limit as N tends to infinity with observations over a fixed time horizon, together with the specific form of the spiking rate and the reset mechanism, must satisfy the technical conditions needed for the classical LAN and minimax theorems to apply.

What would settle it

Numerical simulations in which the variance of the maximum likelihood estimator fails to approach the inverse of the information matrix as N grows large would show that the local asymptotic normality property does not hold.

read the original abstract

In the context of a large system of $N$ neurons interacting through spike events in a mean-field regime as $N\rightarrow \infty$, we characterize the estimation of a multidimensional parameter in the spiking rate, when the neural states are observed over a fixed time horizon. We first prove the local asymptotic normality (LAN) property and leverage classical theory to establish the asymptotic efficiency of the maximum likelihood estimator. While the theory of Ibragimov and Hasminski yields strong results, up to global asymptotic minimax bound, its applicability appears currently limited to models without state resets at spike times. Following then H\"{o}pfner's classical approach, we nevertheless derive, in a general setting including neuron reset, the consistency, asymptotic normality and local asymptotic minimax optimality of the estimator. Keywords: Local Asymptotic Normality (LAN); Mean-field regime; Interacting particle system; Multidimensional parameter estimation; Jump rate estimation; Maximum likelihood estimator (MLE); Asymptotic minimax optimality

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

They prove LAN for the jump-rate parameter in mean-field neuron models with resets and get MLE efficiency via classical methods.

read the letter

They establish local asymptotic normality for the multidimensional parameter governing the jump rate in a mean-field system of neurons, observed over fixed time as the number of neurons goes to infinity. This gives asymptotic efficiency for the maximum likelihood estimator. When the model includes resets of neuron states after spikes, they switch to Höpfner's approach to recover consistency, asymptotic normality, and local asymptotic minimax optimality. The new part is extending the LAN framework to include those resets, which earlier applications of Ibragimov-Hasminski theory could not handle. They work out the details for the limiting McKean-Vlasov process and the associated likelihood ratio. The argument looks solid. The stress test found no internal contradictions or unstated assumptions that would break the passage from finite N to the LAN expansion. They rely on classical external results rather than anything self-referential. A minor point is that everything hinges on the specific spiking rate and reset satisfying the usual regularity conditions like differentiability in quadratic mean. The paper claims to check these in a general setting, but the verification needs to be tight for the jumps. This is for people working on asymptotic statistics for stochastic interacting systems, especially those modeling neural activity. Someone wanting a rigorous efficiency result for MLE in this context would get something out of it. It deserves a serious referee because the result is new for this model class and the technical work appears careful. I would send it out for peer review.

Referee Report

2 major / 2 minor

Summary. The paper establishes the local asymptotic normality (LAN) property for a multidimensional parameter in the spiking rate of a mean-field interacting neuron system (N→∞ limit, fixed time horizon T). It leverages the Ibragimov-Hasminski theory to conclude asymptotic efficiency of the maximum likelihood estimator in the non-reset case. For the version with neuron resets at spike times, it applies Höpfner's classical approach to obtain consistency, asymptotic normality, and local asymptotic minimax optimality of the estimator.

Significance. If the verifications hold, the results supply a rigorous asymptotic foundation for efficient parameter estimation in mean-field limits of jump processes arising in neuroscience. The work correctly identifies the need for a separate treatment of the reset dynamics and invokes the appropriate classical theorems, which is a methodological strength when the technical conditions are fully checked for the McKean-Vlasov limit process.

major comments (2)

[LAN derivation (non-reset case)] The section deriving the LAN property for the non-reset case must explicitly verify differentiability in quadratic mean and the contiguity condition for the limiting process; without these calculations shown in detail, applicability of the Ibragimov-Hasminski theorem cannot be confirmed from the given outline.
[Höpfner approach for reset dynamics] In the reset case, the application of Höpfner's theorem requires a clear statement that the uniform integrability and local asymptotic minimax conditions hold for the mean-field empirical measure; the current sketch leaves open whether the reset mechanism introduces additional bias terms that must be controlled.

minor comments (2)

Notation for the multidimensional parameter θ and the intensity function λ should be introduced once and used consistently; several passages switch between vector and component-wise notation without warning.
The abstract and introduction cite the classical theorems but omit the precise statements of the regularity conditions being checked; adding a short table or list of verified conditions would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading of the manuscript and the constructive comments on the technical conditions needed to apply the classical theorems. We will revise the paper to provide the explicit verifications requested, thereby strengthening the justification for the LAN property and the application of Höpfner's approach.

read point-by-point responses

Referee: [LAN derivation (non-reset case)] The section deriving the LAN property for the non-reset case must explicitly verify differentiability in quadratic mean and the contiguity condition for the limiting process; without these calculations shown in detail, applicability of the Ibragimov-Hasminski theorem cannot be confirmed from the given outline.

Authors: We agree that a complete application of the Ibragimov-Hasminski theorem requires explicit verification of differentiability in quadratic mean and the contiguity condition with respect to the limiting McKean-Vlasov process. In the revised manuscript we will expand the relevant section to include these detailed calculations, confirming that the LAN property holds for the multidimensional spiking-rate parameter in the non-reset case. revision: yes
Referee: [Höpfner approach for reset dynamics] In the reset case, the application of Höpfner's theorem requires a clear statement that the uniform integrability and local asymptotic minimax conditions hold for the mean-field empirical measure; the current sketch leaves open whether the reset mechanism introduces additional bias terms that must be controlled.

Authors: We appreciate the referee's observation. Although the manuscript invokes Höpfner's framework in a setting that already incorporates resets, we acknowledge that uniform integrability of the empirical measure and control of any reset-induced bias terms need to be stated explicitly. In the revision we will add a dedicated paragraph verifying these conditions for the mean-field limit and confirming that no additional bias affects the local asymptotic minimax optimality. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation applies external classical theorems to model-specific verifications

full rationale

The paper proves the LAN property for the multidimensional jump-rate parameter in the N→∞ mean-field limit by verifying standard regularity conditions (differentiability in quadratic mean, contiguity) on the limiting McKean-Vlasov process and its likelihood. For the reset case it checks the hypotheses of Höpfner's theorem to obtain consistency, asymptotic normality and local asymptotic minimax optimality. These steps consist of explicit calculations for the interacting particle system and do not reduce any claimed prediction or optimality result to a quantity defined by the fitted parameter itself. No self-citation chain is load-bearing; the cited results are external classical theorems whose applicability is checked rather than assumed by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard stochastic process theory and classical asymptotic statistics results applied to the mean-field limit; no free parameters or invented entities are introduced.

axioms (1)

standard math Standard assumptions of probability theory and stochastic processes for defining the interacting particle system and its mean-field limit
Invoked to set up the model and apply LAN theory.

pith-pipeline@v0.9.0 · 5478 in / 1164 out tokens · 42052 ms · 2026-05-07T12:32:16.055400+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

36 extracted references · 16 canonical work pages

[1]

Amorino, I

C. Amorino, I. Nourdin, R. Shevchenko 2025 Fractional interacting particle system: drift parameter estimation via Malliavin calculus, Stochastic Processes and Applications; pp 1-38. https://arxiv.org/pdf/2502.06514

work page arXiv 2025
[2]

Amorino, V

C. Amorino, V. Pilipauskaite. (2024) Kinetic interacting particle system: parameter estimation from complete and partial discrete observations, Stochastic Processes and Applications; under revision. https://arxiv.org/pdf/2410.10226

work page arXiv 2024
[3]

Amorino, D

C. Amorino, D. Belomestny, V. Pilipauskaite, M Podolskij, S. Zhou. (2024) Polynomial rates via deconvolution for nonparametric estimation in McKean-Vlasov SDEs, Probability Theory and Related Fields; pp. 1-46

2024
[4]

Amorino, A

C. Amorino, A. Heidari, V. Pilipauskaite, M. Podolskij. (2023) Parameter estimation of discretely observed interacting particle systems, Stochastic Processes and Applications; pp. 350-386

2023
[5]

Applebaum, D. (2009). L \'e vy Processes and Stochastic Calculus, Ed. 2. Cambridge University Press. 2512800

2009
[6]

Billingsley, P. (1999). Convergence of probability measures. John Wiley & Sons

1999
[7]

R., Bryant, H

Brillinger, D. R., Bryant, H. L. and Segundo, J. P. , Identification of synaptic interactions, Biol. Cybern. 22(4) (1976), 213--228, https://doi.org/10.1007/BF00365087

work page doi:10.1007/bf00365087 1976
[8]

, Statistics of spike trains in conductance-based neural networks: Rigorous results, J

Cessac, B. , Statistics of spike trains in conductance-based neural networks: Rigorous results, J. Math. Neurosci. 1 (2011) 1--42, https://doi.org/10.1186/2190-8567-1-8

work page doi:10.1186/2190-8567-1-8 2011
[9]

and Ost, G., A model for neuronal activity in the absence of external stimuli, Markov Process

Duarte, A. and Ost, G., A model for neuronal activity in the absence of external stimuli, Markov Process. Related. Fields 22 (2016), 37-52, http://math-mprf.org/journal/articles/id1412/

2016
[10]

and Laxa, K

Duarte, A. and Laxa, K. and Löcherbach, E. and Loukianova,D., Nonparametric estimation of the jump rate in mean field interacting systems of neurons. https://arxiv.org/abs/2506.24065

work page arXiv
[11]

and L\"ocherbach, E., Infinite systems of interacting chains with memory of variable length--a stochastic model for biological neural nets, J

Galves, A. and L\"ocherbach, E., Infinite systems of interacting chains with memory of variable length--a stochastic model for biological neural nets, J. Stat. Phys. 151(5) (2013), 896--921, https://doi.org/10.1007/s10955-013-0733-9

work page doi:10.1007/s10955-013-0733-9 2013
[12]

Heidari and M

A. Heidari and M. Podolskij (2025): Local asymptotic normality for discretely observed McKean-Vlasov diffusions

2025
[13]

, Mean-field limit of generalized H awkes processes, Stoch

Chevallier, J. , Mean-field limit of generalized H awkes processes, Stoch. Proc. Appl 127 (2017), 3870--3912, https://doi.org/10.1016/j.spa.2017.02.012

work page doi:10.1016/j.spa.2017.02.012 2017
[14]

and Ost, G., Mean field limits for nonlinear spatially extended Hawkes processes with exponential memory kernels, Stoch

Chevallier, J., Duarte, A., L\"ocherbach, E. and Ost, G., Mean field limits for nonlinear spatially extended Hawkes processes with exponential memory kernels, Stoch. Proc. Appl. 129 (2019), 1--27, https://doi.org/10.1016/j.spa.2018.02.007

work page doi:10.1016/j.spa.2018.02.007 2019
[15]

and Pouzat, C

Galves, A., L\"ocherbach, E. and Pouzat, C. , Probabilistic Spiking Neuronal Nets -- Neuromathematics for the Computer Era , Lecture Notes on Mathematical Modelling in the Life Sciences, Springer, Heidelberg, Germany, 2024, https://doi.org/10.1007/978-3-031-68409-8

work page doi:10.1007/978-3-031-68409-8 2024
[16]

2025 Nonparametric moment method for scalar McKean-Vlasov stochastic differential equations ESAIM Probability and Statistics 29

Genon-Catalot, V., Larédo C., Comte F. 2025 Nonparametric moment method for scalar McKean-Vlasov stochastic differential equations ESAIM Probability and Statistics 29

2025
[17]

and Veltz, R., Long time behavior of a mean-field model of interacting neurons, Stoch

Cormier, Q., Tanr\'e, E. and Veltz, R., Long time behavior of a mean-field model of interacting neurons, Stoch. Proc. Appl . 130 (2020), 2553-2595, https://doi.org/10.1016/j.spa.2019.07.010

work page doi:10.1016/j.spa.2019.07.010 2020
[18]

and Veltz, R., Hopf bifurcation in a Mean-Field model of spiking neurons, Electron

Cormier, Q., Tanr\'e, E. and Veltz, R., Hopf bifurcation in a Mean-Field model of spiking neurons, Electron. J. Probab. 26 (2021), 1--40, https://doi.org/10.1214/21-EJP688

work page doi:10.1214/21-ejp688 2021
[19]

and Hoffmann, M., Hawkes processes on large networks, Ann

Delattre, S., Fournier, N. and Hoffmann, M., Hawkes processes on large networks, Ann. App. Probab. 26 (2016), 216--261, https://doi.org/10.1214/14-AAP1089

work page doi:10.1214/14-aap1089 2016
[20]

and Fournier, N., Statistical inference versus mean field limit for H awkes processes, Electron

Delattre, S. and Fournier, N., Statistical inference versus mean field limit for H awkes processes, Electron. J. Stat. 10 (2016), 1223--1295, https://doi.org/10.1214/16-EJS1142

work page doi:10.1214/16-ejs1142 2016
[21]

and Hoffmann, M

Della Maestra, L. and Hoffmann, M. (2023). The LAN property for McKean-Vlasov models in a mean-field regime. Stochastic Process. Appl. 155 109--146

2023
[22]

and Presutti, E

De Masi, A., Galves, A., L \"o cherbach, E. and Presutti, E. (2015). Hydrodynamic Limit for Interacting Neurons. Journal of Statistical Physics 158, 866--902

2015
[23]

Erny, X. (2023). Well-posedness and propagation of chaos for McKean–Vlasov equations with jumps and locally Lipschitz coefficients. Stoch. Proc. Appl. 150, 192--214

2023
[24]

and L \"o cherbach, E., On a toy model of interacting neurons, Ann

Fournier, N. and L \"o cherbach, E., On a toy model of interacting neurons, Ann. Inst. Henri Poincar\' e Probab. Stat. , 52 (2016),1844--1876, https://doi.org/10.1214/15-AIHP701

work page doi:10.1214/15-aihp701 2016
[25]

and Guillin, A

Fournier, N. and Guillin, A. (2015). On the rate of convergence in Wasserstein distance of the empirical measure. Probab. Theory Relat. Fields 162, 707--738

2015
[26]

and L \"o cherbach, E

Fournier, N. and L \"o cherbach, E. (2016). On a toy model of interacting neurons. Ann. Inst. H. Poincar \'e Probab. Statist. 52(), 1844--1876

2016
[27]

H \'a jek, J. (1972). Local asymptotic minimax and admissibility in estimation. of the Sixth Berkeley Symposium on Mathematical Statistics and Probability 1 , 175--194. Univ. California Press

1972
[28]

Ibragimov, I.A., & Hasminski, R. Z. (2013). Statistical estimation: asymptotic theory , 16 Springer Science & Business Media,

2013
[29]

Ikeda, N., & Watanabe, S. (1989). Differential Equations and Diffusion Processes, Ed. 2. North-Holland Publishing Company

1989
[30]

and Protter, P

Jacod, J. and Protter, P. (2012). Discretization of processes , 67. Springer Science & Business Media

2012
[31]

and Shiryaev, A

Jacod, J. and Shiryaev, A. (2013). Limit theorems for stochastic processes . Springer Science & Business Media

2013
[32]

H\"opfner, R. (2014). Asymptotic Statistics: With a View to Stochastic Processes . De Gruyter

2014
[33]

and Touboul, J

Robert, P. and Touboul, J. (2016). On the dynamics of random neuronal networks. J. Stat. Phys 165(3), 545--584

2016
[34]

and L\"ocherbach, E., Multi-class oscillating systems of interacting neurons, Stoch

Ditlevsen, S. and L\"ocherbach, E., Multi-class oscillating systems of interacting neurons, Stoch. Proc. Appl. 127 (2017), 1840--1869, https://doi.org/10.1016/j.spa.2016.09.013

work page doi:10.1016/j.spa.2016.09.013 2017
[35]

Sznitman, A. S. (1991). Topics in propagation of chaos. \'E cole d' \'E t \'e de Probabilit \'e s de Saint-Flour XIX - 1989. Springer Berlin Heidelberg

1991
[36]

and Gerstner, W., Towards a theory of cortical columns: From spiking neurons to interacting neural populations of finite size, PLoS Comput

Schwalger, T., Deger, M. and Gerstner, W., Towards a theory of cortical columns: From spiking neurons to interacting neural populations of finite size, PLoS Comput. Biol. 13(4) (2017), e1005507, https://doi.org/10.1371/journal.pcbi.1005507

work page doi:10.1371/journal.pcbi.1005507 2017