3. Appendix to §2
Here we show how local solutions to the system dx/dt = A(t) x on two overlapping intervals can be combined to give a solution on their union, thus filling in the gap in the proof of the Proposition in §2. We in fact get a slightly stronger result, based on the following
Lemma. for any s0 in the domain of A(t), there exists a neighbourhood N of x0 such that for each s ∈ N, the set {x(s)| x is a solution of dx/dt = A(t) x} is equal to Rn.
Proof. Let {c(1), ... , c(n)} be a basis of the vector space Rn. By what had been established already, we can find a neighbourhood N1 of s0 and functions x(i)(t) such that dx(i)/dt = A(t) x(i), x(i)(s0) = c(i), i = 1, ... , n. Let x(i)(t) = (ai1, ... , ain). Then, if
| | a11(t) | a12(t) | ... | a1n(t) | | | |||
D(t) = | | | a21(t) | a22(t) | ... | a2n(t) | | | ||
| | . | . | ... | . | | | |||
| | an1(t) | ah2(t) | ... | ann(t) | | |
D(1) = 1. Hence, by continuity of the determinant function, we can find a neighbourhood N of s0, contained in N1, such that D ≠ 0 on N. Clearly N satisfies the requirements of the lemma.
||||
We now come to the matter of combining solutions on overlapping neighbourhoods. Let N and P be two intervals with nonempty intersection on which the solutions to the system are defined and satisfy the property stated in the lemma. Let r ∈ N ∩ P. Let {x(i)(t)}1 ≤ i ≤ n and {y(j)(t)}1 ≤ j ≤ n be linearly independent solutions of dx/dt = A(t) x on N and P respectively. If
x(i)(t) = ∑1≤j≤naij(t) c(j) and y(i)(t) = ∑1≤j≤nbij(t) c(j)
we let
G(t) = [aij(t)]1≤i,j≤n and H(t) = [bij(t)]1≤i,j≤n.
Clearly G(r) = B H(r) for some nonsingular matrix B = [βij]1≤i,j≤n. Extend the domain of x(i)(t), 1 ≤ i ≤ m to the whole of N ∪ P by setting x(i)(t) = ∑jβij y(j)(t) on P. This gives ua a set of on the union which are linearly independent at each point.