I think you're worrying too much. This is the correct approach (I'm going to be slightly flippant, so don't take this first paragraph too seriously on a first reading :) ):
- Step 1: Understand the meaning of the Picard-Lindelöf Theorem;
- Step 2: Understand that, by assigning state variables to all but the highest order derivative, you can rework $\ddot x +\omega^2\,x=0$ into a vector version of the standard form $\dot{\mathbf{u}} = f(\mathbf{u})$ addressed by the PL theorem and that, in this case, the $f(\mathbf{u})$ fulfills the conditions of the PL theorem (it is Lipschitz continuous)
- Step 3: Choose your favorite method for finding a solution to the DE and boundary conditions - tricks you learn in differential equations 101, trial and error stuffing guesses in and seeing what happens ..... anything! .... and then GO FOR IT!
Okay, that's a bit flippant, but the point is that you know from basic theoretical considerations there must be a solution and, however you solve the equation, if you can find a solution that fits the equation and boundary conditions, you simply must have the correct and only solution no matter how you deduce it.
In particular, the above theoretical considerations hold whether the variables are real or complex, so if you find a solution using complex variables and they fit the real boundary conditions, then the solution must be the same as the one that is to be found by sticking with real variable notation. Indeed, one can define the notions of $\sin$ and $\cos$ through the solutions of $\ddot x +\omega^2\,x=0$ and they have to be equivalent to complex exponential solutions through the PL theorem considerations above. You can then think of this enforced equivalence as the reason for your own beautifully worded insight that you have worked out for yourself:
"So using sin/cos and even is essentially equivalent so long as you allow for complex constants to provide a conversion factor between the two."
Drop the word "essentially" and you've got it all sorted!
Actually, let's go back to the Step 2 in my "tongue in cheek" (but altogether theoretically sound) answer as it shows us how to unite all of these approaches and bring in physics nicely. Break the equation up into a coupled pair of first order equations by writing:
$$\dot{x} = \omega\,v;\, \dot{v} = -\omega\,x$$
and now we can write things succinctly as a matrix equation:
$$\dot{X} = -i\,\omega \, X;\quad i\stackrel{def}{=}\left(\begin{array}{cc}0&-1\\1&0\end{array}\right)\text{ and } X = \left(\begin{array}{c}x\\v\end{array}\right)\tag{1}$$
whose unique solution is the matrix equation $X = \exp(-i\,\omega\,t)\,X(0)$. Here $\exp$ is the matrix exponential. Note also, that as a real co-efficient matrix, $i^2=-\mathrm{id}$. Now, you may know that one perfectly good way to represent complex numbers is the following: the field $(\mathbb{C},\,+,\,\ast)$ is isomorphic to the commutative field of matrices of the form:
$$\left(\begin{array}{cc}x&-y\\y&x\end{array}\right);\quad x,\,y\in\mathbb{R}\tag{2}$$
together with matrix multiplication and addition. For matrices of this special form, matrix multiplication is commutative (although of course it is not generally so) and the isomorphism is exhibited by the bijection
$$z\in\mathbb{C}\;\leftrightarrow\,\left(\begin{array}{cc}\mathrm{Re}(z)&-\mathrm{Im}(z)\\\mathrm{Im}(z)&\mathrm{Re}(z)\end{array}\right)\tag{3}$$
So if, now, we let $Z$ be a $2\times2$ matrix of this form, then we we can solve (1) by mapping the state vector $X = \left(\begin{array}{c}x\\v\end{array}\right)$ bijectively to the $2\times 2$ matrix $Z = \left(\begin{array}{cc}x&-v\\v&x\end{array}\right)$, solving the equation $\dot{Z} = -i\,\omega\,Z$, i.e. $Z(t) = \exp(-i\,\omega\,t)\,Z(0)$, where $Z(0)$ is the $2\times 2$ matrix of the form (2) with the correct values of $x(0)$ and $v(0)$ that fulfill the boundary conditions, and then taking only the first column of the resulting $2\times 2$ matrix solution $Z(t)$ to get $X(t)$.
This is precisely equivalent to the complex notation method you have been using, as I hope you will see if you explore the above a little. The phase angles are encoded by the phase of the $2\times2$ matrix $Z$, thought of as a complex number by the isomorphism described above.
Moreover, there is some lovely physics here. Consider the square norm of the state vector $X$; it is $E = \frac{1}{2}\,\langle X,\,X\rangle = \frac{1}{2}(x^2 + v^2)$ and you can immediate deduce from (1) that
$$\dot{E} = \langle X,\,\dot{X}\rangle = X^T\,\dot{X} = -\omega\,X^T \,i\, X = 0\tag{4}$$
This has two interpretations. Firstly, $E$ is the total energy of the system, partitioned into potential energy $\frac{1}{2}\,x^2$ and kinetic $\frac{1}{2}\,v^2$. Secondly, (4) shows that the state vector, written as Cartesian components, follows the circle $x^2+v^2=2\,E$ and indeed this motion is uniform circular motion of $\omega$ radians per unit time. So that simple harmonic motion is the motion of any Cartesian component of uniform circular motion.
You could also solve the problem by beginning with (1), deducing (4) and then make the substition
$$x=\sqrt{2\,E}\,\cos(\theta(t));\quad\, v=\sqrt{2\,E}\,\sin(\theta(t))\tag{5}$$
which is validated by the conservation law $x^2+v^2=2\,E$ with $\dot{E}=0$. Then substitute $x$ back into the original SHM equation to deduce that
$$\theta(t) = \pm\omega\,t+\theta(0)\tag{6}$$
Best Answer
I think the reason for the sentence
Is that $\eta_i$ represents (likely) the displacement of a particle from a certain position. Something for example measured in meters and therefore, obviously, a real quantity. This is the reason for the "of course" in the sentence as a complex value wouldn't make sense.
This however is a bit cheating. We have a mathematical description of reality. Said description allowed to be augmented to complex values. But if the description is correct it must give back real values for proper (real) data.
And this is indeed what happens. First you should notice that there are actually two solutions
$$ \eta_i = C_\pm e^{\pm i \omega t} $$
Any linear combination of the above is still a solution. The point is that, if we did things correctly, for proper initial conditions (i.e. initial conditions that are compatible with meaningful data) the solution should automatically turns out real.
To summarize, the solution of the equation, being a second order ODE, depends on two initial conditions (for example initial position and velocity). If these values correspond to a meaningful motion, the end solution turns out automatically real.