This post aims to answer a simple question in linear system theory: Is a system with all its poles on the imaginary axis always marginally stable?

The answer is NO. A well-known counter example is (1)x˙=[0100]x The system in (1) has two poles at the origin, but it is not marginally stable since there is a single Jordan block corresponding to the eigenvalue 0.

Now, let us make this question a bit more challenging: Is any system with all its poles satisfying (a) non-zero (b) on the imaginary axis always marginally stable?

This revised question excludes the counter example in (1). According to Wikipedia,

a continuous-time state-space model is marginally stable if and only if the Jordan blocks corresponding to poles with zero real part are one-by-one matrices.

By Wiki (which might be wrong), the answer to the revised question seems negative, since the Jordan blocks corresponding to the non-zero poles on the imaginary axis may not be one-by-one. But I am not convinced by this argument unless someone can show me a counter example.

So, our goal now is to find a real matrix whose the Jordan block corresponding to an imaginary eigenvalue is not one-by-one. Have you seen such an example in your linear system theory course or linear algebra 101? I cannot recall anything like that. One famous example for a matrix with purely imaginary poles is (2)A=[01 10]. The Jordan normal form of A is (3)[i00i]. It would be nice if we can find a real matrix whose Jordan normal form is (4)[i100 0i00 00i1 000i]. Actually, it is indeed possible. According to this document (Real Jordan Form), the matrix in (5) has exactly the Jordan normal form in (4).

(5)A=[0110 1001 0001 0010]. Therefore, the system x˙=Ax with A in (5) serves as a counter example for our revised question.