In perturbation theory, why can perturbed eigenfunctions be expanded into the basis set of the unperturbed eigenfunctions? So I am studying non-degenerate time independent perturbation theory and I came across the derivation of the first order correction to the wavefunction. So the notes given to me affirm that the each "perturbed" part of the wavefunction can be expanded into: 
$$\psi^{(k)}_n=\sum_{k} c^{(k)}_m \psi^{(0)}_m$$ 
where $\psi^{0}_m$ are the unperturbed eigenfunctions. So for example the first order correction to the wavefunction would be:
$$\psi^{(1)}_n=\sum_{m \neq n} c^{(1)}_m \psi^{(0)}_m$$ 
I don't understand why this can be done in the first place, what principle does it follow from? Why can the basis set of the unperturbed eigenfunctions be used to define the perturbed corrections? Also why does $m \neq n$ necessary in this definition (I am aware that the inner product  $$\langle\psi^{(0)}_m|\psi^{(1)}_n \rangle = \frac{\langle\psi^{(0)}_m|H'|\psi^{(1)}_n \rangle}{E^{(0)}_n - E^{(0)}_m}$$ implies that this is valid only for $n \neq m$ as the denominator would be otherwise zero).
 A: The unperturbed solutions form a complete set so that any function can be expanded in terms of functions in this set, just like any function on the sphere can be expanded in terms of $Y_{\ell,m}(\theta,\varphi)$ or any periodic function function on the line can be expanded in terms of $\cos(m x)$ and $\sin (n x)$.
The real issue is the practicality of using the unperturbed solutions: after all, to find some perturbed radial solution, you could choose to expand using the unperturbed radial functions for the spherical oscillator or the Coulomb problem.  Physically however, it makes sense to use the unperturbed solutions: since the perturbed ones ought to be “close” to the unperturbed ones, one can expect the series to converge quickly.
Choosing the unperturbed solutions as a starting point is a bit like asking what is the best starting point to evaluate $\sqrt{17}$: you could start by writing 
$$
\sqrt{17}=\sqrt{16+1}=4\sqrt{1+1/16}\approx 4\left(1+\frac{1}{32}\right)
=4+\frac{1}{8}=4.125\tag{1}
$$
or you would write
$$
\sqrt{17}=\sqrt{25-8}\approx 5\left(1-\frac{8}{50}\right)=5-\frac{8}{10}=4.2
\, .\tag{2}
$$
Obviously (1) is a “better” zero’th order approximation (at least to first order) because $\sqrt{16}$ is clearly a better approximation to $\sqrt{17}$ than would be $\sqrt{25}$.  
A: The eigenfunctions of the unperturbed Hamiltonian form a complete basis, so any function  can be expressed as a linear combination of those eigenfunctions. This is how you obtain your first formula.
Your second formula is not a definition, it is a result of calculation in perturbation theory (see, e.g., here).
