📊 statistics

A Kuramoto-von Mises Time Series Model for Probabilistic Modeling of Coupled Oscillators

This paper proposes a novel Kuramoto-von Mises time series model based on Langevin dynamics to efficiently estimate the probability distributions of coupled oscillators in non-equilibrium regimes, offering a closed-form maximum likelihood solution that outperforms existing equilibrium-based methods in applications ranging from brain traveling waves to gastrointestinal slow wave activity.

Original authors: Yun Hwang, Todd P. Coleman

Published 2026-06-16

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Yun Hwang, Todd P. Coleman

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a world filled with thousands of tiny metronomes, each ticking at its own slightly different speed. In the real world, these aren't just metronomes; they are neurons firing in your brain, cells in your stomach contracting to digest food, or even proteins folding into shape. When these "metronomes" are close to each other, they don't just tick alone; they influence one another, trying to sync up or push each other out of rhythm. This is what scientists call a system of coupled oscillators.

For a long time, scientists trying to map out how these metronomes talk to each other had to make a big, risky guess: they assumed the whole system was "calm and settled" (in equilibrium), like a lake with no wind. They also assumed that if Metronome A influenced Metronome B, then B influenced A exactly the same way (symmetry).

The Problem:
Real life is messy. Systems are often chaotic, windy, and constantly changing. In the brain or the stomach, the influence isn't always equal (A might push B hard, but B barely pushes back), and the system is rarely "settled." The old methods, which relied on that "calm lake" assumption, often failed or gave biased answers when applied to these real, dynamic situations.

The New Solution:
The authors of this paper, Yun Hwang and Todd Coleman, built a new tool called the Kuramoto–von Mises Time Series Model. Think of it as a new kind of detective kit that doesn't need the system to be calm to solve the mystery.

Here is how it works, using simple analogies:

1. The "Control Signal" Analogy

Imagine you are driving a car (Oscillator A) and you see a car ahead of you (Oscillator B) going slightly faster. To stay in sync, you press the gas. The old models assumed that if you pressed the gas to match B, B was doing the exact same thing to match you.
The new model realizes that in the real world, the road might be one-way, or B might be a truck that doesn't care about your car. The new method calculates the "push" and "pull" between every pair of oscillators without assuming they are mirror images of each other. It handles the traffic, not just the calm parking lot.

2. The "High-Speed Camera" Trick

The secret sauce of this new method is how it looks at time.

The Old Way: Tried to guess the rules by looking at the final, settled state of the system.
The New Way: Uses a "high-speed camera" (modern data acquisition systems). If you take pictures of a spinning fan fast enough, you can see exactly how the blades move from one frame to the next.
Because modern sensors record data very quickly (thousands of times a second), the authors realized they could use a mathematical shortcut (a "small angle approximation"). This shortcut turns a very complex, messy math problem into a simple, straight-line algebra problem.
The Result: They found a closed-form solution. In plain English, this means they found a direct formula to calculate the connections. You don't need a supercomputer to run hours of trial-and-error; you can just plug the numbers into a calculator and get the answer instantly.

3. The "Von Mises" Compass

Since these oscillators are phases (like angles on a circle, 0 to 360 degrees), standard bell-curve statistics (Gaussian) don't work well because 0 degrees and 360 degrees are the same spot. The authors used a special circular distribution called the von Mises distribution.
Think of it like a compass. If you are trying to predict where a compass needle will point next, you don't just look at "North" or "South"; you look at the wind (the other oscillators) and the needle's own momentum. The von Mises distribution is the perfect tool for predicting where a needle on a circle will land next, given the wind and its current speed.

What Did They Prove?

The authors tested their new "detective kit" in three main ways:

The Simulation Test: They created fake data where the oscillators were chaotic and influenced each other unevenly (asymmetric). The old methods failed to see the true connections, but the new method saw them clearly. Even when the data was calm and symmetric, the new method worked just as well as the old ones.
The "Traveling Wave" Test: They simulated waves moving across a grid (like a ripple in a pond). They used their model to guess the direction of the wave. The model was excellent at telling the difference between a wave moving North and a wave moving slightly Northeast, proving it can detect subtle changes in how the system moves.
Real-World Tests:
- The Stomach: They looked at electrical recordings from human stomachs. The stomach has waves of activity that help digest food. The model successfully identified that in healthy people, the "connections" between stomach cells get stronger after eating (to help digest), but in people with digestion problems, this connection pattern didn't change.
- The Brain: They looked at brain recordings from a patient receiving Deep Brain Stimulation (DBS). The model showed that during the stimulation, the connections between brain regions weakened, but immediately after the stimulation stopped, the connections "rebounded" and became stronger than before. This helped explain why the brain's activity changed in response to the treatment.

The Bottom Line

This paper presents a faster, more flexible way to map out how things that "wiggle" or "pulse" (like brain cells or stomach muscles) talk to each other. It stops assuming the world is calm and symmetrical, and instead uses high-speed data to calculate the exact rules of the game, even when the game is chaotic. It's a tool that works for both the quiet moments and the stormy ones.

Technical Summary: A Kuramoto–von Mises Time Series Model for Probabilistic Modeling of Coupled Oscillators

Problem Statement
Coupled oscillators provide a fundamental framework for modeling physical and biological phenomena, ranging from protein folding and neural synchronization to gastrointestinal slow-wave activity. Existing methods for estimating the probability distributions of multivariate phase relationships in these systems typically rely on equilibrium thermodynamics, expressing the system via a Boltzmann-form pairwise exponential family distribution. However, these approaches depend on two critical assumptions that are frequently violated in real-world systems: (i) the coupling matrix is symmetric, and (ii) the system has reached thermodynamic equilibrium. Many biological networks exhibit inherently asymmetric interactions (e.g., excitatory/inhibitory synaptic connections) and are driven by external inputs, placing them in non-equilibrium regimes. Consequently, equilibrium-based formulations may yield limited or biased characterizations of underlying dynamics.

Methodology
The authors propose a method for estimating underlying phase distributions without assuming symmetric connectivity or an equilibrium state. The approach is built upon the following theoretical and algorithmic foundations:

Stochastic Kuramoto Formulation: The model extends the deterministic Kuramoto model to a stochastic differential equation (SDE) using Langevin dynamics. This formulation incorporates temporal delays and models the continuous-time evolution of oscillator phases $Y_t$ with additive Gaussian noise.
Von Mises Approximation: By discretizing time into small intervals $\Delta$ , the conditional distribution of the phase at time $t+\Delta$ given the history is approximated. Since the phase is circular ( $[0, 2\pi]$ ), the wrapped Gaussian distribution resulting from the SDE is approximated by a von Mises distribution. The mean direction of this distribution is determined by the intrinsic frequency and the coupling terms, while the concentration parameter is inversely proportional to the noise variance and time step.
High Sampling Rate Approximation: A key simplification is applied under the assumption of high sampling rates ( $\Delta \ll 1$ ). Using Taylor approximations for sine and cosine functions where the coupling influence is small, the log-likelihood function is linearized. This transforms the maximum likelihood estimation (MLE) problem into a closed-form algebraic solution involving linear algebra operations, avoiding the need for iterative convex optimization solvers.
Parameter Estimation: The method estimates coupling parameters ( $\kappa_{ij}$ ) and phase delays ( $\mu_{ij}$ ) by solving a separable quadratic optimization problem for each oscillator. The authors also provide a procedure to estimate the noise variance $\sigma^2$ and demonstrate how to enforce symmetric connectivity constraints if desired, though the general framework does not require it.

Key Contributions

Non-Equilibrium Modeling: The paper introduces a probabilistic framework that does not require the system to be in thermodynamic equilibrium or possess symmetric coupling, addressing a significant limitation of prior Boltzmann-based approaches.
Closed-Form Solution: The authors derive a maximum likelihood estimator that admits a closed-form algebraic solution in the high sampling rate regime, making the method computationally efficient and readily applicable to modern high-frequency data acquisition systems.
Unified Framework: The approach unifies the modeling of both equilibrium and non-equilibrium regimes within a single Langevin-based formulation.

Results
The authors evaluate the method using both synthetic data and real-world electrophysiological recordings:

Synthetic Graph Recovery: In simulations of symmetric, equilibrium systems, the proposed method recovers connectivity structures with accuracy comparable to the state-of-the-art ISO method (which assumes equilibrium). In asymmetric, non-equilibrium systems, the proposed method significantly outperforms the ISO method, which fails to capture the asymmetric structure, while standard linear regression performs poorly across all dimensions.
Traveling Wave Classification: The method is applied to simulated 2D planar traveling waves. Using log-likelihood ratio tests, the model successfully classifies waves propagating in different directions, with classification performance (AUC) improving as the difference in propagation direction increases.
Gastric Electrogastrography (EGG): Applied to human stomach recordings, the model detects changes in gastric traveling waves before and after meals. In healthy controls, the model identifies increased local connectivity during digestive periods. In patients with gastroparesis or functional dyspepsia, the model captures a pathological lack of change in connectivity between pre- and post-meal states.
Intracranial EEG: The method is applied to deep brain stimulation (DBS) data from the orbitofrontal cortex. It captures significant changes in network connectivity during and after stimulation, observing an attenuation of connectivity during stimulation and a rebound increase afterward, consistent with observed changes in theta power.

Significance and Claims
The paper claims that the proposed Kuramoto–von Mises time series model offers a robust and generalizable alternative to existing equilibrium-based methods. By leveraging the high sampling rates common in modern data acquisition, the method provides a computationally efficient, closed-form solution for estimating coupling structures in both equilibrium and non-equilibrium regimes. The authors emphasize that while the method relies on a small-angle approximation (justified by high sampling rates relative to intrinsic frequencies), it successfully characterizes dynamic phenomena such as traveling waves and pathological connectivity patterns in biological systems where traditional equilibrium assumptions fail. The work suggests that this framework enables scalable and interpretable analysis of high-dimensional coupled oscillator systems without the restrictive assumptions of symmetry or equilibrium.

1. The "Control Signal" Analogy

2. The "High-Speed Camera" Trick

3. The "Von Mises" Compass

What Did They Prove?

The Bottom Line

More like this