5.3 Stability

Definition 5.3.1. The zero solution of \(x' = Ax\) is stable if every solution remains bounded as \(t \to \infty \) (equivalently, for linear systems, solutions starting close to \(0\) stay close), and asymptotically stable if every solution tends to \(0\).

Theorem 5.3.2. Let \(\sigma = \max \{\Real \lambda : \lambda \text { an eigenvalue of } A\}\), the spectral abscissa.

(i)
For every \(\mu > \sigma \) there is a constant \(K\) with \(\|e^{At}\| \leq Ke^{\mu t}\) for \(t \geq 0\).
(ii)
The zero solution is asymptotically stable if and only if \(\sigma < 0\), and then every solution decays exponentially.
(iii)
If \(\sigma > 0\), some solutions are unbounded.
(iv)
If \(\sigma = 0\), the zero solution is stable if and only if every eigenvalue with zero real part has Jordan blocks of size one.

Proof. (i) By Theorem 5.2.1, every entry of \(e^{Jt}\) has modulus at most \(t^ke^{\sigma t}/k!\) for some \(k < n\), and \(t^ke^{\sigma t} \leq C_ke^{\mu t}\) for \(t \geq 0\) because \(t^ke^{-(\mu - \sigma )t}\) is bounded. Then \(\|e^{At}\| \leq \|P\|\|e^{Jt}\|\|P^{-1}\| \leq Ke^{\mu t}\). (ii) If \(\sigma < 0\), take \(\mu = \sigma /2 < 0\) in (i). Conversely if \(\lambda \) is an eigenvalue with \(\Real \lambda \geq 0\) and eigenvector \(v\), then \(e^{\lambda t}v\) (or its real part) is a solution that does not tend to \(0\). (iii) The same solution grows like \(e^{(\Real \lambda )t}\). (iv) If an eigenvalue \(ib\) has only blocks of size one, its terms in \(e^{Jt}\) are \(e^{ibt}\), bounded; the other eigenvalues have negative real parts and contribute decaying terms. If it has a block of size \(m \geq 2\), the solution with initial value the second vector of the block’s Jordan chain contains \(te^{ibt}\), which is unbounded. □

Theorem 5.3.2(i) is sharp about rates but silent about constants. For a non-normal matrix the constant \(K\) can be large, and solutions of an asymptotically stable system can grow substantially before they decay (Software 5.6.3). This transient growth matters in fluid mechanics, where it is a route to turbulence in flows that are linearly stable, and in ecology and control.

Questions on this section

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