Solvable macroscopic models for populations of type I excitability neurons with Gaussian noise
summary
The gist
Solvable macroscopic models for populations of type I excitability neurons with Gaussian noise provides an analytical solution for the first passage time problem in a quadratic integrate-and-fire…
In short
This work develops solvable macroscopic models for type I excitability neurons modeled as quadratic integrate-and-fire (QIF) neurons subjected to white Gaussian noise. It provides analytical solutions for the first passage time problem, allowing calculation of firing rates and population membrane potentials in both homogeneous and heterogeneous populations. The results extend to include networks with electrical synapses, linking QIF statistics to Anderson localization in quantum mechanics.
Key concepts
- Quadratic Integrate-and-Fire (QIF) Neuron
- This is a mathematical model describing the dynamics of type I excitability neurons near a bifurcation point. It models the neuron's membrane potential and input current, specifically focusing on how it fires when stimulated by noise.
- First Passage Time Problem (FPTP)
- This problem seeks to find the time it takes for a stochastic process, like the membrane potential of a QIF neuron, to reach a specific threshold. Solving this yields analytical expressions for key population metrics like firing rate and mean interspike interval.
- Montbri´o–Paz´o–Roxin (MPR) Order Parameter
- This is a dimensionless quantity, W = πr - iv, used to characterize the macroscopic activity of the neural population. It combines the firing rate (r) and the population-mean membrane potential (v), providing a single measure of how synchronized or active the network state is.
Terminology used across episodes
This episode discusses
- Solvable macroscopic models for populations of type I excitability neurons with Gaussian noise · Paper Radio
The paper
Solvable macroscopic models for populations of type I excitability neurons with Gaussian noise · Read on arXiv
Maria V. Ageeva, Denis S. Goldobin
Institute of Continuous Media Mechanics, Ural Branch of RAS · Institute of Physics and Mathematics, Perm State University · Department of Control Theory, Lobachevsky State University of Nizhny Novgorod
The dynamics of type I excitability neurons near the bifurcation point are governed by the model of quadratic integrate-and-fire neuron (QIF). The model of QIF subject to white Gaussian noise is abundant in statistical physics, since it not only reasonably reproduces the dynamics of some biological systems but also emerges in the problems of stochastic stability and in quantum mechanics of one-dimensional systems with quenched disorder. We derive the solution for the mean interspike interval of a QIF subject to additive Gaussian noise in terms of analytic functions, from the first passage time problem. On the basis of this solution, we construct analytical expressions for the firing rate and population-mean membrane potential in terms of the real and imaginary parts of a certain universal analytic function. With this function, we generalize theoretical results to heterogeneous populations with Lorentzian-distributed parameters. Implementation of theoretical results is demonstrated and examined for two problems: the interplay between chemical and electrical synapses, and the emergence of effective endogenous noise in sparse inhibitory networks. For the latter case, it is possible to cover the scenarios where the effective noise intensity depends on Lorentzian-distributed parameters, which is not feasible directly with the diffusion approximation.
Transcript
Introduction to the show: ident: Condensed Matter. The day in condensed matter: materials, phases and the measurements under them.
Kai: I'm Kai.
Mira: And I'm Mira, here with Lev.
Kai: We're looking at the paper: "Solvable macroscopic models for populations of type I excitability neurons with Gaussian noise".
Mira: Here's the gist of the paper. Solvable macroscopic models for populations of type I excitability neurons with Gaussian noise provides an analytical solution for the first passage time problem in a quadratic integrate-and-fire neuron subject to white Gaussian noise, which allows for the calculation of firing rate and population-mean membrane potential in homogeneous and heterogeneous populations Model Formulation The dynamics of type I excitability neurons near the bifurcation point are governed by the model of quadratic integrate-and-fire neuron (QIF) subject to white Gaussian noise The QIF model is defined by the differential equations for membrane potential and synaptic input current The probability density function (PDF) PA(V, t) for a homogeneous population follows the Fokker–Planck equation The primary quantifiers of mean-field activity are the firing rate and the population-mean membrane potential Homogeneous Population Analysis The solution to the first passage time problem (FPTP) yields an analytical expression for the mean interspike interval, which is then used to calculate the firing rate r and population-mean membrane potential v The Montbri´o–Paz´o–Roxin (MPR) order parameter W = πr − iv can be calculated as W = D1/3W(A), where W(A) is given by a specific expression involving Bessel functions The dimensionless mean interspike interval is given by hτi = F(0) = π2
Ai(-A): squared +
Bi(-A): squared Heterogeneous Populations and Generalization The analytical result can be extended to cases involving Lorentzian-distributed parameters, allowing generalization to heterogeneous populations with Lorentzian-distributed parameters The population mean values of the firing rate and membrane potential are determined by the weighted integral over real A using the function hWi = R +∞−∞ W(A)LA(A) dA The final result for the order parameter is given by W = πr − iv = D1/3W(A), where W(A) is defined in terms of Bessel functions Mean-Field Network Applications The mean-field impact of electrical synapses on QIF networks is equivalent to a mere decrease of the chemical synapse heterogeneity For time-independent states, the population firing rate and mean potential can be determined by solving quartic equations parameterized by v and r The system of equations for the macroscopic state is given by (50) Network Dynamics with Noise For sparse inhibitory networks, the applicability of the effective noise model and corresponding mean-field theory for networks with parameter heterogeneities and node in-degree K heterogeneity has been validated through finite network simulations The results from network simulations confirm the validity of this approach The statistical theory problem of QIFs is mathematically equivalent to the problem of Anderson localization in one-dimensional Schr¨odinger equation with a frozen random potential The presented solutions for time-independent macroscopic regimes are also applicable for quasi-stationary (adiabatically changing) network configurations, which is practically useful for theoretical studies of neural plasticity The study confirms that the impact of electrical synapses is more sophisticated than in the D = 0 case The mathematical equivalence between QIF statistics and Anderson localization provides a deep connection to quantum mechanics problems The paper concludes by showing how the Residue theorem can be applied to the diffusion-approximation model with Lorentzian-distributed intensity Dj The results are also applicable for other parameter distributions
33: Pruning redundant reasoning tokens The dynamics of type I excitability neurons near the bifurcation point are governed by the model of quadratic integrate-and-fire neuron (QIF) The model of QIF subject to white Gaussian noise is abundant in statistical physics, since it not only reasonably reproduces the dynamics of some biological systems but also emerges in the problems of stochastic stability and in quantum mechanics of one-dimensional systems with quenched disorder How the training signal is built The firing rate r can be expressed via hτi (e.g.,
27: ): r = M T1(−∞) + T2(−∞) + · · · + TM(−∞) = D1/3 hτi The population-mean membrane potential v is calculated using the integral (25), which leads to v = D1/3V(A) and V(A) being expressed in terms of Bessel functions Where the gains come from The asymptotic laws for r and v are given by (35) and (36), which describe the behavior of the firing rate and population-mean potential as A approaches infinity The dependence of hri and hvi on A0 is shown in Fig.
Kai: 3, illustrating the dependencies on A0 for heterogeneous populations The solution (54) provides a self-consistent system for macroscopic states parameterized by A0 and ∆A The solvability condition for the linear equations defining the boundary of the hysteretic transition domain is given by (51) The final result for the order parameter is summarized in (33) and (34) Pruning redundant reasoning tokens The paper is organized as follows.
Mira: In Sec.
Kai: II, we introduce the QIF model and its Fokker–Planck equation For a homogeneous population of QIFs, we derive the mean interspike interval as an analytical solution to the first passage time problem in Sec.
Title and authors: Kai: Okay, let's talk about the title and the people behind this work, "Solvable macroscopic models for populations of type I excitability neurons with Gaussian noise." The authors are Maria V. Ageeva and Denis S. Goldobin from institutes in Perm Russia.
Mira: They’re bringing expertise in condensed matter theory, so they’re focused on the underlying assumptions that make this solvable—specifically, how the white Gaussian noise interacts with the QIF dynamics to yield these analytical solutions.
Lev: I'm interested in what they actually built and measured here. If we take these mean-field results, what is the practical overhead? Does calculating this order parameter W and getting those scaling laws really translate to running a real network of QIFs on superconducting qubits?
Kai: Well, the paper focuses on the time-independent states for now, which is a good starting point because it simplifies things significantly before we even deal with time evolution. They are showing that under these conditions, you can find those firing rate and mean potential equations by solving some quartic equations parameterized by the value of A and delta A.
Mira: That specific algebraic self-consistency problem they set up using those quartic equations is a key improvement because it allows them to tackle more complex scenarios than just the simplest cases. It lets them explore how different noise intensities or parameter spreads affect the population state.
The paper's summary: Kai: So, summarizing what they found, this paper gives us an analytical expression for the mean interspike interval of a QIF neuron under white Gaussian noise by solving the first passage time problem in Fourier space. That leads directly to the firing rate r and the population mean membrane potential v.
Mira: The key result is that these quantities are expressed in terms of universal analytic functions, specifically involving Airy functions and then being related to Bessel functions of order one third. This is what allows them to generalize the solution from a homogeneous population to one with Lorentzian parameter distributions.
Lev: That connection between Airy and Bessel functions is mathematically beautiful, but for me, I need to know how stable this analytical structure is when we introduce the noise intensity D. The paper gives an expression for W, which is that Montbri´o–Paz´o–Roxin order parameter, as W being proportional to D times a function of A.
Kai: That proportionality to D shows how the noise directly scales the resulting firing rate and potential. It confirms that the noise level isn't just a small perturbation; it fundamentally changes the macroscopic behavior in a predictable way.
The paper's improvements: Mira: One of the main improvements they highlight is extending this analytical result to cases involving Lorentzian distributions for parameters, which opens up a whole new avenue for studying real-world biological variability.
Kai: They show how to calculate the population mean values of the firing rate and membrane potential by performing a weighted integral over that distribution using the function h Wi, which is defined using that Montbri´o–Paz´o–Roxin order parameter formula.
Lev: That integral approach is powerful, but it means we have to solve an algebraic self-consistency problem for those mean-field states parameterized by A zero and delta A, which adds a layer of complexity for any simulation you might try to build on this.
Kai: They also show how the mean-field impact of electrical synapses on QIF networks is equivalent to just a decrease in chemical synapse heterogeneity, which simplifies the analysis somewhat when we look at those time-independent states.
Conclusion: Mira: So, to wrap up this paper, these solutions for the time-independent macroscopic regimes are applicable not just for static systems but also for quasi-stationary or adiabatically changing network configurations, which is useful for studying neural plasticity over time.
Kai: It really confirms that the impact of electrical synapses is more sophisticated than in the simplest cases, and they show a deep connection between QIF statistics and Anderson localization in one-dimensional Schrödinger equations with a frozen random potential.
Lev: I think that connection to quantum mechanics is what gives this result its real power. If we can map these population dynamics onto such a known quantum problem, it suggests there are universal principles at play that might be applicable beyond the specific QIF model and noise intensity we used here.
Mira: Agreed, and I think the way they handle parameter distributions by using that integral over real A to find the final order parameter is a really solid way to move toward a self-consistent description of heterogeneous systems.
Kai: So, in short, this paper provides an analytical toolkit for calculating population firing rates and potentials in QIFs under Gaussian noise, and they show how you can handle real biological complexity like heterogeneity by extending the math to include Lorentzian parameter distributions.
Lev: For those of us building hardware, it’s a solid benchmark because it shows what the theoretical framework looks like before we have to worry about the massive computational cost of simulating these complex systems directly.
Mira: Exactly, and we're looking forward to seeing how this framework for solving self-consistent problems can be applied to more dynamic or higher-dimensional neural models next.
More episodes
- 2610.32678-Probing the limits of materials design via a comprehensive crystal database of 14.6 million structures with 140 properties
- 2610.33595-Native-oxide-driven layer asymmetry, two-step switching and exchange bias in Co/Ir/Co trilayers
- 2503.19572-Spin models from nonlinear cellular automata
- 2606.23350-Irreversibility Enhances Quantum-Enhanced Markov-Chain Monte Carlo
- 2608.02744-Two-Parameter Ansatz for the Violation of Eigenstate Thermalization
- 2210.00806-Low- and high-energy localization landscapes for tight-binding Hamiltonians in two-dimensional lattices
- 2409.09625-Selective switch for band-edge alignments in pentagonal CdSeTe monolayer: Atom-valley locking
- 2606.20168-Norms, overlaps and Yangian descendants for the Haldane-Shastry spin chain
- 2512.19503-Strain-induced splitting of the CCDW-NCCDW phase transition in 1T-TaS 2
- 2307.12638-Strong localization of microwaves beyond two dimensions in aperiodic Vogel spirals