为何雅可比矩阵的迹与特征值可判定平衡点稳定性？

Hey there! As someone who’s wrestled with this exact question back when I was learning differential equations, let’s break this down in super concrete terms—no topology or Lyapunov functions required, promise.

First: What’s the Jacobian Doing Here? #

Lotka-Volterra equations are nonlinear (those x*y terms make the system behave like a curve), but right near an equilibrium point, the system acts almost exactly like a linear equation. Think of zooming way in on a curved line: it starts to look straight. The Jacobian matrix is just the "straightened-out" version of the nonlinear system at that equilibrium. It tells us how tiny changes in prey/predator numbers will grow or shrink near that balance point.

Why Eigenvalues Matter #

For linear systems (which our zoomed-in Jacobian represents), solutions follow the form e^(λt) * vector, where λ is an eigenvalue. Here’s what that means for stability:

• If λ has a negative real part: e^(λt) gets smaller and smaller over time. Any tiny deviation from the equilibrium will fade away—so the equilibrium is stable.
• If λ has a positive real part: e^(λt) blows up as time passes. Tiny deviations grow into big shifts away from the equilibrium—so it’s unstable.
• If λ is pure imaginary (no real part): e^(λt) just oscillates (like sine/cosine). Deviations don’t grow or shrink—this is neutral stability (think the classic periodic predator-prey cycles in Lotka-Volterra).

The Trace: A Quick Stability Shortcut #

The trace of the Jacobian is just the sum of its diagonal entries. But here’s a key linear algebra fact you might already know: the trace equals the sum of all eigenvalues. For 2D systems (like Lotka-Volterra, with prey and predator variables), this gives us a fast check:

• Trace < 0: The sum of eigenvalues is negative. Either both eigenvalues are negative real numbers, or they’re complex with negative real parts. Either way, deviations fade—stable equilibrium.
• Trace > 0: The sum of eigenvalues is positive. At least one eigenvalue has a positive real part, so deviations grow—unstable equilibrium.
• Trace = 0: The sum of eigenvalues is zero. Now we need to look at the determinant (product of eigenvalues):
  • Determinant > 0: Eigenvalues are pure imaginary (sum to 0, product positive). This is neutral stability (the periodic cycles you see in Lotka-Volterra’s coexistence equilibrium).
  • Determinant < 0: Eigenvalues are one positive, one negative real number. This is a saddle point—unstable, since deviations will grow in one direction.

Let’s Tie This to Lotka-Volterra #

Take the standard predator-prey model:

dx/dt = x(r - py)  # Prey growth (r = intrinsic rate, p = predation rate)
dy/dt = y(-m + qx) # Predator growth (m = death rate, q = feeding efficiency)

We have two key equilibria to analyze:

1. (0, 0) (no prey, no predators):
   • Jacobian: [[r, 0], [0, -m]]
   • Eigenvalues: r (positive) and -m (negative)
   • Trace = r - m (could be positive or negative, but since one eigenvalue is positive, this equilibrium is unstable—makes sense: add a tiny number of prey, and they’ll grow exponentially!)
2. (m/q, r/p) (coexisting prey and predators):
   • Jacobian: [[0, -pm/q], [qr/p, 0]]
   • Trace = 0, Determinant = m*r (positive)
   • Eigenvalues: ±i√(mr) (pure imaginary)
   • This is neutral stability—populations cycle around the equilibrium instead of settling to it, which matches what you see in Lotka-Volterra simulations.

Wrap-Up #

The Jacobian lets us use linear math (which you already know!) to guess how the nonlinear system behaves near equilibria. Eigenvalues tell us exactly how deviations change over time, and the trace is a quick way to sum up that behavior without calculating every eigenvalue. It’s all about zooming in, simplifying, and applying the linear rules you already understand.

内容的提问来源于stack exchange，提问作者George Orf.