System of first order linear ODEs (real case)

It is a particular case of the linear complex case and, more in general, of the system of first order ODEs.
If A is a real matrix, given a solution φ(t), the conjugate function is just the solution of the system whose initial value is the conjugate of that of φ(t).
By uniqueness of solutions, if we take any real initial condition, vRn then

eAtv

must be real, and so eAt is a real matrix.

What form does the elements of this matrix have? In the note of the complex case we stated that

eAt=MeJtM1

Solutions are then made by mixing quasi-polynomials by mean of complex number, so the results are linear combinations of real and imaginary parts of these quasi-polynomials:

Proposition

Let z˙=Az a system of first order linear equations, with A:RnRn a linear operator. Then, if A has real eigenvalues λi with algebraic multiplicities ni and complex eigenvalues αj±iωj with multiplicities μj then every component of a solution φ(t) has the form:

φm(t)=ieλitpm,i+jeαjt[qm,j(t)cosωjt+rm,j(t)sinωjt]

where pm,i(t), qm,j(t) and rm,j(t) are real polynomials with degree limited by the multiplicities.

In particular, consider the case where all eigenvalues are single. We would have:

(1)φm(t)=icm,ieλit+jam,jeαjtcosωjt+bm,jeαjtsinωjt

with a,b,c families of real constants. But there are too many.

How many of degrees of freedom do we have? Let's see.
The map φ:IRCn is a complex solution of the equation if and only if the real and the imaginary part of z are both real solutions. This is because of linearity and because z(t) is also a solution of the associated complex equation (the only one with initial value the complex conjugate of z(0)).

Suppose A has n distinct eigenvalues, r real eigenvalues and c complex eigenvalues, such that n=r+2c. The general solution is

φ(t)=k=1nckeλktξk

where we have chosen real eigenvectors for the real eigenvalues, and complex conjugate eigenvectors for the complex eigenvalue pairs. We observe that for φ(0) being real

k=1nckξk=k=1nckξk

and therefore coefficients of complex conjugates eigenvectors must be complex conjugates. So, for real eigenvalues (with real eigenvector) ck is real, and we denote it by ak. And for the complex conjugates eigenvalues λk and λj, with of course complex conjugates eigenvectors, coefficients ck and cj must be conjugates. So we can write the solution as

φ(t)=k=1rakeλktξk+k=r+1r+cckeλktξk+k=r+1r+cckeλktξk

and so

φ(t)=k=1rakeλktξk+k=r+1r+c2Re[ckeλktξk]

In conclusion, in equation (1) above, cm,i, am,j, bm,j are "connected" in such a way that they only suppose n degrees of freedom: they are linear functions of real and imaginary parts of ck and we have that the first r cks are real and the last 2c are paired by conjugation .
Sometimes it is preferred other expression of the general solution. For k=r+1..r+c the terms above can be rewritten as:

2Re[ckeλktξk]=Re(ξk)eαkt(2Re(ck)cos(ωkt)2Im(ck)Sin(ωkt))++Im(ξk)eαkt(2Im(ck)cos(ωkt)2Re(ck)Sin(ωkt))+

And this can be cleaned be means of a change of names and trigonometric properties into:

eαktAkcos(ωkt+ϕk)vk1+eαktAksin(ωkt+ϕk)vk2,

where vk1=Re(ξk), vk2=Im(ξk), Ak=|2ck| and ϕk=arg(2ck))