Linearization and the Jacobian Matrix

Core Idea

Near a fixed point x*, a nonlinear system ẋ = f(x) can be approximated by its linearization ẋ ≈ Df(x*) · (x - x*), where Df(x*) is the Jacobian matrix of partial derivatives evaluated at the fixed point. The Hartman-Grobman theorem guarantees that this linear approximation captures the correct qualitative behavior — the nonlinear flow is topologically equivalent to the linearized flow — provided all eigenvalues have nonzero real parts (the fixed point is hyperbolic).

Explainer

Nonlinear systems are, in general, impossible to solve exactly. But near a fixed point, the nonlinear terms are small compared to the linear ones (because the state variables are close to zero after shifting coordinates to place the fixed point at the origin). This is the fundamental insight behind linearization: replace the hard problem with an easy one that's accurate where it matters most — in the neighborhood of equilibrium.

The tool is the Jacobian matrix Df(x*), the matrix of all first partial derivatives of f evaluated at the fixed point. If ẋ = f(x) and x* is a fixed point with f(x*) = 0, then Taylor-expanding around x* gives ẋ = Df(x*)(x - x*) + higher-order terms. Dropping the higher-order terms yields the linearized system, which is just a linear ODE whose solution you know from your work on eigenvalues: the eigenvalues and eigenvectors of Df(x*) completely determine the local dynamics. Negative real parts mean attraction, positive mean repulsion, imaginary parts mean oscillation.

The Hartman-Grobman theorem makes this precise. It says that if x* is a hyperbolic fixed point (all eigenvalues of Df(x*) have nonzero real parts), then there exists a continuous change of coordinates that maps the nonlinear flow onto the linear flow near x*. The phase portraits are topologically equivalent — they have the same qualitative structure. A stable node stays a stable node. A saddle stays a saddle. An unstable spiral stays an unstable spiral. The nonlinear terms can warp trajectories, change speeds, and distort shapes, but they cannot change the topology of the local flow.

The theorem's restriction to hyperbolic fixed points is not a technicality — it's where all the interesting dynamics hide. When an eigenvalue has zero real part, the linearization is on a knife's edge: the nonlinear terms determine whether the system tips toward stability or instability. A linear center (purely imaginary eigenvalues) might become a stable spiral, an unstable spiral, or remain a center in the nonlinear system. A zero eigenvalue often signals a bifurcation — a qualitative change in the number or stability of fixed points as a parameter varies. The Hartman-Grobman theorem tells you exactly where linearization succeeds and, equally importantly, where you need more sophisticated tools.