Revealing dynamics of non-autonomous complex systems… — Plain-Language Explanation

Imagine you are trying to figure out the secret recipe for a complex dish, like a stew, but you can only taste the final soup. You know the ingredients (the vegetables, spices, and meat) change over time, and you suspect that the heat of the stove and the amount of water added are the hidden forces driving the flavor. However, you don't have a thermometer to measure the heat, and you can't see the water level. You just have the taste of the soup at every minute.

This is the challenge scientists face when trying to understand complex systems—from how fish swim in a school to how a drone flies or how a heart beats. They have data on the system's behavior, but they often lack the "forcing parameters" (the hidden knobs like temperature, wind, or drug dosage) that drive the changes.

This paper introduces a clever new tool to solve this mystery. Here is how it works, broken down into simple concepts:

1. The Problem: The "Missing Knob"

Most existing tools for finding these hidden recipes assume you know exactly how the "knobs" (external factors) are turning.

The Old Way: If you are studying a plant, you might try to guess the equation by looking at the plant and the sunlight data. But what if you don't have the sunlight data? Or what if the sunlight data is messy and hard to measure? The old tools fail because they are stuck waiting for that missing knob.
The Limitation: Even if you try to guess the knob is just "time," it often leads to a wrong recipe because the relationship isn't that simple.

2. The Solution: The "Magic Proxy"

The authors discovered a mathematical trick. They proved that you don't actually need to know the real value of the hidden knob (like the exact temperature). You only need to know its direction (is it going up or down?).

The Analogy: Imagine you are trying to drive a car up a hill, but you can't see the road or the speedometer. You only know if you are pressing the gas (going up) or the brake (going down).
The authors created a "Magic Proxy" (let's call it Variable $\nu$ ). This is a fake, made-up number that simply goes up or down, mimicking the direction of the real hidden knob.
The Big Discovery: They proved mathematically that if you use this "Magic Proxy" to build your recipe, you get the exact same result as if you had used the real, perfect data of the hidden knob. It's like realizing that you can bake the perfect cake using a timer that just counts "up" and "down," even if you don't know the exact temperature of the oven.

3. The Process: Finding the "Sweet Spot"

Since there are infinite ways to set up this "Magic Proxy" (you can start it at 0 or 100, and make it step up by 1 or 0.001), the computer has to find the best version.

The team built a "search grid" (like a giant spreadsheet) to test thousands of different starting points and step sizes.
They used a special scoring system (called $\epsilon$ AIC) that acts like a judge. This judge doesn't just look at how well the recipe fits the data; it also checks if the math is "clean" and free of calculation errors.
The winner is the version of the "Magic Proxy" that gives the most accurate, stable, and simple equation.

4. What They Tested It On

The team tested this "Magic Proxy" method on four very different real-world scenarios to prove it works:

The Leaf Cell (Energy Crisis): They looked at data from plant cells running out of oxygen. The hidden knob was the dropping oxygen level. Their method successfully predicted the exact moment the cell's energy would suddenly crash (a "tipping point"), even without knowing the oxygen levels.
The Drone (Autonomous Flight): They analyzed a drone flying through obstacles. The hidden knob was the changing environment the drone "saw." The method figured out the drone's control algorithm just by watching its flight path, effectively reverse-engineering the drone's brain.
The Chick Heart (Arrhythmia): They studied heart cells treated with a drug that causes irregular beats. The hidden knob was the drug spreading through the tissue. The method predicted exactly when the heart would switch from a steady beat to a chaotic one.
The Fish Community (Ocean Ecosystem): They looked at 14 species of fish in a bay. The hidden knob was the water temperature changing with the seasons. The method successfully predicted the population booms and busts of the fish, revealing the hidden rules of their survival.

5. Why This Matters

Think of this method as a universal translator for complex systems.

Before: Scientists needed a perfect map of every external factor (temperature, wind, drug dose) to understand a system. If the map was missing, they were stuck.
Now: They only need to know the direction of change (is the factor getting better or worse?) and the behavior of the system itself.

The paper claims this approach allows us to uncover the "laws of nature" governing complex systems—even when those systems are messy, changing, and driven by factors we cannot directly measure. It turns a "black box" into a transparent window, revealing the hidden equations that drive our world.

Technical Summary: Revealing Dynamics of Non-Autonomous Complex Systems from Data

Problem Statement
Discovering governing equations from observational data is a fundamental challenge in understanding complex systems across science and engineering. While recent advances have enabled the discovery of equations for autonomous systems (where dynamics depend solely on state variables), a significant gap remains in handling non-autonomous systems. Real-world systems are predominantly non-autonomous, driven by external, time-varying parameters (forcing parameters) that are often unobservable or inaccessible.

Existing methodologies for non-autonomous systems face two primary limitations:

Unobservability of Forcing Parameters: Standard approaches require the time series of external forcing parameters to construct basis functions. In many real-world scenarios (e.g., ecological or biological systems), these parameters cannot be measured directly.
Heuristic and Costly Library Search: Current methods attempt to infer dynamics by searching for optimal linear combinations within a pre-defined library of basis functions (including state variables, forcing parameters, and time). This approach relies heavily on expert knowledge to define the library, incurs high computational costs due to the vast search space, and risks missing the true governing equation if the library lacks sufficient completeness. Furthermore, simply replacing forcing parameters with a generic time variable ( $t$ ) has not been systematically validated and can lead to biased inference.

Methodology
The authors propose a novel, data-driven framework that shifts the focus from selecting coefficients within a fixed library to adaptively constructing the basis functions themselves. The core of the method relies on a theoretical equivalence theorem and an optimization strategy for numerical errors.

Equivalence Theorem for Basis Functions:
The authors prove that within a given model space, the dynamical equation driven by external forcing parameters $\Phi$ can be equivalently represented by a set of basis functions driven by a single variable $\nu$ .
- Let the forcing parameter evolve as $\Phi_{i+1} = \Phi_i + s_i \Delta\Phi$ , where $s_i \in \{+1, -1\}$ indicates the direction of change.
- The theorem states that a variable $\nu$ , evolving as $\nu_{i+1} = \nu_i + s_i \Delta\nu$ (with arbitrary initial value $\nu_1$ and step size $\Delta\nu$ ), spans the exact same model space as $\Phi$ .
- Consequently, the governing equation can be inferred using $\nu$ without requiring observation of $\Phi$ , provided the sign sequence $\{s_i\}$ (indicating the trend of the forcing parameter) is known or assumed.
Optimal Driving Variable Identification:
While the theorem guarantees theoretical equivalence for any $\nu_1$ and $\Delta\nu$ , practical implementation using pseudo-inverse operations (to solve for coefficients) introduces numerical errors that vary depending on the choice of $\nu$ .
- The framework employs a grid search over candidate pairs of $(\nu_1, \Delta\nu)$ .
- For each pair, the authors solve for the coefficient matrix and evaluate the model using a numerical error adjusted Akaike's Information Criterion ( $\epsilon$ AIC).
- The $\epsilon$ AIC metric explicitly incorporates the normalized numerical error ( $\epsilon$ ) arising from the pseudo-inverse calculation, balancing fitting accuracy, model complexity, and numerical stability.
- The pair $(\nu_1, \Delta\nu)$ that minimizes $\epsilon$ AIC is selected as the optimal driving variable, yielding the most reliable set of basis functions.
Inference Process:
- Construct a feature matrix $\Theta_\nu$ using the state variables and the optimal $\nu$ .
- Solve the linear system $\dot{X} = \Theta_\nu A_\nu$ (using sparse regression like LASSO or ridge regression) to identify the governing equation.
- This process allows for the reconstruction of dynamics even when the true forcing parameters are unknown, provided the qualitative trend (sign sequence) of the driving force is available.

Key Results
The method was validated on both synthetic and empirical systems:

Synthetic Systems:
- Cusp Bifurcation: The method successfully recovered the governing equation and coefficients for a system driven by two bifurcation parameters, outperforming methods using forcing parameters or time variables, even as the basis function library size increased.
- Coupled Kuramoto Oscillators: The approach accurately inferred the coupling equations for a network of oscillators driven by a single parameter (coupling strength), demonstrating robustness across various network sizes.
- Robustness to Library Deficiencies: Crucially, when key functional terms were deliberately removed from the basis library (making the "true" equation unrecoverable by standard sparse regression), the proposed method still generated surrogate models that accurately captured the system's bifurcation behavior and pre-bifurcation trajectories.
Empirical Systems (Real-World Applications):
- Cellular Energy (Leaf Cells): The method inferred the dynamics of ATP concentration collapse under hypoxia. The inferred equation revealed a fold bifurcation, providing a mechanistic explanation for the sudden energy collapse.
- UAV Autonomous Flight: From trajectory data of a drone avoiding obstacles, the method distilled the underlying control dynamics. It accurately predicted future flight paths and revealed the thrust variation patterns (including deceleration phases) without prior knowledge of the control algorithm.
- Chick-Heart Aggregates: Analyzing inter-beat intervals under drug treatment (E-4031), the method identified period-doubling bifurcations, explaining the onset of arrhythmia.
- Marine Fish Community: Using population abundance data from Maizuru Bay, the method inferred species-specific dynamics driven by water temperature. The inferred equations accurately predicted seasonal population fluctuations and captured the system's dynamic stability patterns, aligning with ecological theory.

Significance and Claims
The paper claims to offer a novel paradigm for revealing the underlying dynamics of non-autonomous complex systems. Its primary significance lies in:

Overcoming Unobservability: It enables the inference of governing equations without requiring direct observation of external forcing parameters, a common bottleneck in real-world applications.
Adaptive Basis Construction: Instead of relying on heuristic, expert-defined libraries, the method adaptively identifies an optimal set of basis functions within the model space, reducing computational costs and improving robustness against library incompleteness.
Numerical Stability: By explicitly optimizing for numerical errors via the $\epsilon$ AIC metric, the framework ensures that the inferred equations are not only mathematically consistent but also numerically reliable.
Mechanistic Insight: The method successfully distills complex behaviors (such as bifurcations and control strategies) from raw data, offering a pathway to understand the governing laws of diverse systems ranging from cellular biology to autonomous robotics and ecology.

The authors emphasize that while the method requires a qualitative understanding of the forcing parameter's trend (the sign sequence $s_i$ ), this is a minimal prior specification compared to the full time-series data required by existing methods. The approach is presented as a versatile tool for uncovering the "hidden" dynamics of real-world systems where traditional modeling is infeasible.

Revealing dynamics of non-autonomous complex systems from data