ODE: Nonlinear Systems and Linearization

Nonlinear systems of ordinary differential equations (ODEs) govern phenomena from electrical circuits to ecological populations, but their behavior is rarely solvable in closed form. The most powerful analytical tool we have is to understand their dynamics near critical resting points, known as equilibria. This process of linearization—approximating the complex nonlinear system with a simpler linear one near an equilibrium—is fundamental to predicting stability and classifying local behavior in engineering design, control theory, and mathematical biology.

Finding Equilibria and the Jacobian Matrix

The first step in analyzing any autonomous system $\dot{\mathbf{x}}=\mathbf{f}(\mathbf{x})$, where $\mathbf{x}\in {\mathbb{R}}^n$, is to find its equilibrium points (or fixed points). These are states where the system does not change, found by solving the vector equation $\mathbf{f}({\mathbf{x}}^{\ast })=0$. For a two-dimensional system $\dot{x}=f(x,y),\dot{y}=g(x,y)$, this means simultaneously solving $f(x,y)=0$ and $g(x,y)=0$. A system can have zero, one, or multiple equilibria.

Once an equilibrium ${\mathbf{x}}^{\ast }$ is located, we analyze the local dynamics by examining the system's linear approximation. This approximation is derived from the Jacobian matrix, denoted $J(\mathbf{x})$ or $D\mathbf{f}(\mathbf{x})$. The Jacobian is the matrix of all first-order partial derivatives of the vector field $\mathbf{f}$. For a 2D system, it is computed as:

$$J(x,y)=\left[\begin{array}{cc}\frac{\partial f}{\partial x}& \frac{\partial f}{\partial y}\\ \frac{\partial g}{\partial x}& \frac{\partial g}{\partial y}\end{array}\right]$$

You evaluate this matrix at the equilibrium point to obtain the Jacobian at equilibrium, $J({\mathbf{x}}^{\ast })$. This matrix $J({\mathbf{x}}^{\ast })$ encapsulates the best linear approximation of $\mathbf{f}$ near ${\mathbf{x}}^{\ast }$.

Linearization and the Hartman-Grobman Theorem

Linearization about an equilibrium is the process of constructing the associated linear system $\dot{\mathbf{u}}=J({\mathbf{x}}^{\ast })\mathbf{u}$, where $\mathbf{u}=\mathbf{x}-{\mathbf{x}}^{\ast }$ is the small deviation from equilibrium. This linear system is much easier to analyze: its solution is determined by the eigenvalues and eigenvectors of $J({\mathbf{x}}^{\ast })$.

The profound connection between the nonlinear system and its linearization is formalized by the Hartman-Grobman theorem. This theorem states that if the equilibrium ${\mathbf{x}}^{\ast }$ is hyperbolic—meaning all eigenvalues of $J({\mathbf{x}}^{\ast })$ have non-zero real parts—then the phase portrait of the nonlinear system near ${\mathbf{x}}^{\ast }$ is topologically equivalent to the phase portrait of its linearization. In practical terms, the stability type (node, saddle, spiral) and stability (stable/unstable) predicted by the linear system are guaranteed to hold for the full nonlinear system in a small neighborhood of the equilibrium. This justifies using the eigenvalues of the Jacobian to classify local stability.

Limitations at Non-Hyperbolic Points

The power of linearization hinges on hyperbolicity. Linearization at non-hyperbolic points, where at least one eigenvalue has a zero real part, fails to provide a definitive answer. The Hartman-Grobman theorem does not apply, and the nonlinear terms, which were neglected in the approximation, become decisive in determining stability and local phase portrait structure.

Consider a system with pure imaginary eigenvalues (e.g., a linear center). The linearization predicts neutral cycles, but the nonlinear system could actually have a stable spiral, an unstable spiral, or a true center, depending on higher-order terms. Analyzing such cases requires more advanced techniques like Lyapunov functions or center manifold theory. In engineering applications, a non-hyperbolic equilibrium often indicates a delicate bifurcation point, where small parameter changes can qualitatively alter the system's behavior.

Applications to Predator-Prey and Competition Models

These techniques are beautifully illustrated in biological systems. The classic Lotka-Volterra predator-prey model is given by: $$\dot{x}=\alpha x-\beta xy,\dot{y}=\delta xy-\gamma y$$ where $x$ is prey and $y$ is predator. The non-zero interior equilibrium is found at $(x^{\ast },y^{\ast })=(\gamma /\delta ,\alpha /\beta )$. Computing the Jacobian at this point yields eigenvalues that are pure imaginary, making it a non-hyperbolic (center) case for the linearization. The full nonlinear system, however, conserves a quantity and produces closed orbits—neutral cycles. This shows the limitation: linearization predicts a center, but cannot confirm it; nonlinear analysis is required.

In contrast, competition models (e.g., two species competing for a resource) often have multiple hyperbolic equilibria. For a model like: $${\dot{N}}_1=r_1{N}_1\left(1-\frac{N_1+{\alpha }_{12}{N}_2}{K_1}\right),{\dot{N}}_2=r_2{N}_2\left(1-\frac{N_2+{\alpha }_{21}{N}_1}{K_2}\right)$$ you find equilibria corresponding to extinction of one or both species, or coexistence. Linearizing at each equilibrium and computing eigenvalues allows you to definitively classify their stability. This analysis predicts the model's outcomes: competitive exclusion of one species or stable coexistence, which are critical insights for ecology and resource management.

Common Pitfalls

1. Forgetting to check the hyperbolicity condition: Always compute the eigenvalues of the Jacobian. If you find an eigenvalue with $\text{Re}(\lambda )=0$, you cannot conclude stability from linearization alone. Stating "the equilibrium is a center" based solely on the linearization is a common error.
2. Misapplying linearization far from equilibrium: Linearization is a local approximation. It is only valid in a (possibly very small) neighborhood of the equilibrium point. Using it to make predictions about global, long-term behavior will lead to incorrect conclusions.
3. Algebraic mistakes in Jacobian evaluation: The most frequent computational errors occur in taking partial derivatives or in evaluating them at the equilibrium coordinates. After computing $J(x,y)$, systematically substitute $x=x^{\ast }$ and $y=y^{\ast }$. A sign error in a single derivative can completely change the eigenvalue calculation and your stability conclusion.
4. Overlooking equilibria: Systems can have multiple equilibria. Solving $\mathbf{f}(\mathbf{x})=0$ might yield several solutions, each of which requires independent analysis. Failing to find all equilibria means you have an incomplete picture of the system's possible steady states.

Summary

• The analysis of nonlinear ODE systems begins by finding equilibrium points where the vector field vanishes, $\mathbf{f}({\mathbf{x}}^{\ast })=0$.
• The Jacobian matrix of first partial derivatives, evaluated at the equilibrium, provides the best linear approximation of the system's flow nearby.
• The Hartman-Grobman theorem validates linearization for hyperbolic equilibria (all eigenvalues have non-zero real part), guaranteeing the nonlinear system's local phase portrait matches its linearization's.
• At non-hyperbolic points (where an eigenvalue has zero real part), linear analysis fails, and nonlinear terms determine stability, requiring more advanced methods.
• These tools are essential for analyzing classic biological models like predator-prey and competition models, allowing prediction of stability, coexistence, or extinction outcomes from model parameters.