Right. The entanglement correlates the results of the two sides. But there is only *one* wave function and that wave function works for the entire apparatus.
Next, *every* detector collapses that wave function to some degree, but potentially not all the way. So, a measurement can give a definite value at one point, be correlated, but still not give a definite value at another point.
Next, if you look at only the D0 detector, and don't worry about the results in the others, the result is that you get an interference pattern. The 'paradox' comes when you correlate the results at D0 and those at the other detectors.
So, you get a detection at D0. What can still happen? Well, we don't have 'which slit' information, so we only know there was a detection at D0. Nothing else. The 'collapsed' wave function after that detection is *still* a superposition of wave functions from the two slits.
Now, detectors D3 and D4 *only* detect from a single slit. So, if they are hit, the component of the wave function that is relevant is from that slit and that slit only. Those detectors collapsed the wave function to give that information. And that means that when we *select out* those detections at D0 that correspond to detections at D3 or D4, we won't see an interference pattern: the component of the wave function that would produce it is selected away *after the experiment is run*.
On the other hand, detections at D2 do NOT have 'which slit' information, which means that the component of the wave function after that still has component from both slits and so will show an interference pattern. The D2 collapse is to an interference state.
One crucial thing here is that, just like in the Aspect experiment, things at D0 look random: there is an interference pattern. It is only when you put the information together that you see these correlations. Everything happens in a forward time direction and through propagation of probabilities and correlations through the wave function. Furthermore, each measurement 'collapses' the wave function to some degree, but not completely.
Right. QM is a non-causal, but local theory. it is also not a realist theory: particles do NOT have definite properties at all times: they have *probabilities* of different values for properties. And those probabilities can be *correlated* by entanglement.
The wave functions propagate locally, but the value of any measurement is undetermined ahead of time, although *correlations* are. The correlation between the two sides of the apparatus is made at that initial prism that splits the beams and is propagated to both sides. When a measurement is made, we know some *part* of the wave function, but not all of it.