10.3 Hartman–Grobman and stable manifolds

For a hyperbolic equilibrium, linearisation describes not just stability but the whole local phase portrait.

Theorem 10.3.1 (Hartman–Grobman). If \(x^*\) is a hyperbolic equilibrium, there are neighbourhoods \(V\) of \(x^*\) and \(W\) of \(0\) and a homeomorphism \(h\colon V \to W\) that maps orbits of \(x' = f(x)\) in \(V\) to orbits of the linearisation \(u' = Au\) in \(W\), preserving the direction of time: \(h(\phi _t(x)) = e^{At}h(x)\) while both sides remain in the neighbourhoods.

We do not prove this theorem; the standard proof constructs \(h\) as the fixed point of a contraction on a space of continuous maps. The conjugacy is only continuous in general, not differentiable, so it preserves the qualitative picture (a saddle stays a saddle, a node a node, though a node and a focus are topologically equivalent) but not quantities such as angles. Smoother information is carried by the invariant manifolds.

Theorem 10.3.2 (Stable manifold theorem). Let \(x^*\) be a hyperbolic equilibrium, with \(k\) eigenvalues of negative real part and \(n - k\) of positive real part, and let \(E^s\), \(E^u\) be the corresponding (generalised) eigenspaces. Then near \(x^*\) there are \(C^1\) manifolds \(W^s_{\mathrm {loc}}\) of dimension \(k\) and \(W^u_{\mathrm {loc}}\) of dimension \(n - k\), tangent at \(x^*\) to \(E^s\) and \(E^u\), such that \begin {gather*} W^s_{\mathrm {loc}} = \{x \text { near } x^* : \phi _t(x) \to x^* \text { as } t \to +\infty \},\\ W^u_{\mathrm {loc}} = \{x \text { near } x^* : \phi _t(x) \to x^* \text { as } t \to -\infty \} . \end {gather*}

The stable manifold of a saddle in the plane is a pair of orbits approaching the saddle from opposite sides, and it is the boundary between regions of different behaviour: a separatrix. It can be computed by starting very close to the saddle along the stable eigenvector and integrating backwards in time, which is how Figure 14 was drawn.

Example 10.3.3. For \(x' = x\), \(y' = -y + x^2\), the origin is a saddle with \(E^u\) the \(x\)-axis and \(E^s\) the \(y\)-axis. The \(y\)-axis is invariant (if \(x = 0\) it stays \(0\)), so \(W^s = \{x = 0\}\). For the unstable manifold try \(y = cx^2\): along it \(y' = 2cxx' = 2cx^2\), while the equation requires \(y' = -cx^2 + x^2\), so \(2c = 1 - c\) and \(c = \frac 13\). The parabola \(y = x^2/3\) is invariant and tangent to the \(x\)-axis at the origin: it is \(W^u\), curved away from its linear approximation \(E^u\) by the non-linear term.

Questions on this section

Stuck on something here? Ask below and it stays attached to this topic.