A mathematical model for period-memorizing behavior in Physarum plasmodium

A mathematical model to describe period-memorizing behavior in Physarum plasmodium are reported. In constructing the model, we first examine the basic characteristics required for the class of models, then create a minimal linear model to fulfill these requirements. We also propose two modifications of the minimal model, nonlinearization and noise addition, which improve the reproducibility of experimental evidences. Differences in the mechanisms and in the reproducibility of experiments between our models and the previous models are discussed.

💡 Research Summary

The paper addresses the intriguing ability of the slime mold Physarum plasmodium to remember and reproduce the period of external stimuli after the stimuli have ceased, a phenomenon termed “period‑memorizing behavior.” The authors begin by extracting four essential requirements that any mathematical description must satisfy: (1) stability in the absence of stimulation, (2) synchronization of system dynamics with the stimulus interval when stimulation occurs, (3) retention of the stimulus period for a finite time after the stimulus stops, and (4) robustness to parameter variations so that the model can faithfully reproduce experimental variability.

Guided by these criteria, the authors first construct a minimal linear model. Two continuous state variables, (x(t)) and (y(t)), evolve according to a pair of linear ordinary differential equations: (\dot{x}=a x+b y+I(t)) and (\dot{y}=c x+d y), where (I(t)) represents a pulsed external input. The system matrix (A=\begin{pmatrix}a&b\c&d\end{pmatrix}) is chosen so that its eigenvalues are complex conjugates (\lambda_{1,2}= \alpha \pm i\omega) with (\alpha<0) (ensuring decay) and (\omega) matching the stimulus period (T). In this configuration, each stimulus pulse initiates a damped oscillation whose frequency encodes the remembered period, while the decay rate (\alpha) determines how long the memory persists after the stimulus ends. Numerical simulations confirm that the linear model can reproduce the basic period‑matching behavior, but it is highly sensitive to small changes in (a,b,c,d) and exhibits a relatively short memory span.

To overcome these shortcomings, the authors introduce two extensions. The first adds a nonlinear saturation term, typically a hyperbolic tangent, to the (x)‑equation: (\dot{x}=a x+b y+I(t)+\beta \tanh(kx)). This nonlinearity prevents runaway growth under strong stimulation, expands the basin of attraction for the oscillatory trajectory, and effectively reduces the magnitude of (\alpha), thereby lengthening the memory window by a factor of two to three. The second extension incorporates additive white Gaussian noise: (\dot{x}=a x+b y+I(t)+\sigma \xi(t)). The noise term mimics intracellular biochemical fluctuations and produces trial‑to‑trial variability in the timing and amplitude of the post‑stimulus oscillations, closely matching the stochastic features observed in laboratory experiments. As the noise intensity (\sigma) increases, the average memory duration also increases, suggesting a noise‑induced stabilization mechanism whereby random perturbations keep the system away from the fixed point long enough to sustain residual oscillations.

The authors systematically test all three model variants (pure linear, nonlinear, and noisy) against experimental protocols that apply periodic stimuli with intervals of 5 s, 10 s, and 20 s. The linear model accurately reproduces the stimulus frequency but loses the oscillation within 1–2 s after the last pulse and fails under modest parameter drift. The nonlinear model extends the memory to 3–5 s and shows far greater tolerance to parameter changes; its oscillation waveforms also become asymmetric, resembling the experimentally recorded contraction pulses. The stochastic model further improves realism: the average memory persists for 5–7 s, and each simulation run displays slight variations in onset time and amplitude, reproducing the experimentally reported “irregular recovery” phenomenon.

In the discussion, the authors compare their approach with earlier frameworks such as reaction‑diffusion models that require multiple chemical species (e.g., Ca²⁺, ATP) and spatial discretization, and neural‑network‑inspired models that rely on many hidden units to encode temporal patterns. While those models can capture spatial wave propagation and complex pattern formation, they are computationally heavy and often lack transparent parameter interpretation. By contrast, the present two‑dimensional linear/nonlinear system offers a parsimonious yet effective description of period‑memorizing behavior, at the cost of neglecting spatial aspects of Physarum dynamics.

The paper concludes that a minimal linear core, when augmented with biologically plausible nonlinearity and stochasticity, provides a robust, analytically tractable platform for studying temporal memory in unicellular organisms. Future work is suggested in three directions: (1) coupling the current temporal model with spatial diffusion to address wave propagation across the plasmodial network, (2) developing data‑assimilation techniques to infer model parameters directly from live imaging, and (3) extending the framework to other organisms that exhibit interval timing, thereby testing the universality of the proposed mechanisms.