I have found the following exercise in one of my problem sheets:
Suppose we have an observable $Q$ and its corresponding operator $\hat{Q}$ has three eigenfunctions $\varphi_1, \varphi_2, \varphi_3$ with eigenvalues $2, 2,$ and $0$, respectively. Let $\psi$ be the following superposition state: $$\psi(t=0) = \varphi_1 + \frac{1}{\sqrt{2}}\varphi_2 + i\varphi_3$$ If a precise measurement of $Q$ yields the value $2$, what will be the wavefunction immediately after the measurement?
My guess was that $\psi$ must now be a superposition of the eigenstates $\varphi_1$ and $\varphi_2$, since these are the only ones with eigenvalue 2 for the observable $Q$. That is, immediately after the measurement, $\psi = a\varphi_1 + b\varphi_2$, for some $a, b \in \mathbb{C}$ (up to normalisation). However, the solution turns out to be that $\psi = \varphi_1 + \frac{1}{\sqrt{2}}\varphi_2$ up to a normalising constant. I do not understand why the coefficients $1$ and $\frac{1}{\sqrt{2}}$ are "preserved". Wouldn't any arbitrary linear combination of $\varphi_1$ and $\varphi_2$ be possible? Doesn't the state $\psi$ collapse into the eigenspace spanned by $\varphi_1$ and $\varphi_2$?