7.2 Floquet’s theorem

Lemma 7.2.1. Every invertible complex matrix \(C\) has a logarithm: a matrix \(B\) with \(e^B = C\).

Proof. By Theorem 5.1.2(iv) it suffices to treat a Jordan block \(C = \lambda I + N = \lambda (I + N/\lambda )\) with \(\lambda \neq 0\) and \(N^m = 0\). Choose a complex logarithm \(\log \lambda \) and set \[ B = (\log \lambda )I + L, \qquad L = \sum _{k=1}^{m-1}\frac {(-1)^{k+1}}{k}\Bigl (\frac N\lambda \Bigr )^k , \] the series for \(\log (1 + x)\) with \(x = N/\lambda \), which terminates because \(N\) is nilpotent. The formal power series identity \(\exp (\log (1 + x)) = 1 + x\) holds as an identity of polynomials in \(x\) modulo \(x^m\), and substituting the nilpotent matrix \(N/\lambda \) gives \(e^L = I + N/\lambda \). Since \((\log \lambda )I\) commutes with \(L\), \(e^B = \lambda e^L = C\). □

Even when \(C\) is real, \(B\) may have to be complex: a real logarithm exists if and only if the negative real eigenvalues of \(C\) have Jordan blocks occurring in pairs. However, \(C^2\) always has a real logarithm, which is why the real form of Floquet’s theorem needs period \(2T\).

Theorem 7.2.2 (Floquet). There are a constant matrix \(B\), possibly complex, and a \(T\)-periodic invertible matrix function \(P(t)\) such that \[ \Phi (t) = P(t)e^{Bt} . \] The change of variables \(x = P(t)y\) transforms \(x' = A(t)x\) into the constant-coefficient system \(y' = By\). If \(A\) is real, one can choose \(B\) real and \(P\) real and \(2T\)-periodic.

Proof. By Lemma 7.2.1 choose \(B\) with \(e^{BT} = M\), and define \(P(t) = \Phi (t)e^{-Bt}\), which is invertible. Then \(P(t + T) = \Phi (t + T)e^{-Bt}e^{-BT} = \Phi (t)Me^{-BT}e^{-Bt} = \Phi (t)e^{-Bt} = P(t)\), using \(e^{-BT}e^{-Bt} = e^{-Bt}e^{-BT}\). Every solution is \(x = \Phi (t)c = P(t)e^{Bt}c\), so \(y = P(t)^{-1}x = e^{Bt}c\) solves \(y' = By\), and conversely every solution \(e^{Bt}c\) of \(y' = By\) gives a solution \(x = P(t)y\). For the real form, \(M^2\) is real and has a real logarithm \(\tilde B\) with \(e^{2T\tilde B} = M^2\); then \(P(t) = \Phi (t)e^{-\tilde Bt}\) is real and \(2T\)-periodic by the same calculation. □

Every solution is therefore a periodic matrix times a solution of a constant-coefficient system: a periodic modulation of exponential behaviour. Floquet’s theorem does not give \(B\) or \(P\) explicitly, since computing \(M\) requires solving the equation over one period, but it reduces every question about long-run behaviour to the eigenvalues of \(M\).

Questions on this section

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