Some time ago I read some stuff about delayed choice and quantum eraser experiments, but rather than reading the articles thorouhgly, I just glanced at them quickly, picked up a few ideas, and then tried to design a gedanken experiment of my own. Now I seem to have run into a bit of a problem. My gedanken experiment seems to either contain a paradox or violate the principles of quantum mechanics. I'm sure I'm making some sort of mistake, but so far I haven't been able to find it. Perhaps someone here can can help me figure this out. Look at the attached drawing. A "down converter" is a crystal that that emits two photons, each with half the energy of the original, when it's hit with just one. According to what I read, such crystals actually exist. I imagine that it would be difficult, if not practically impossible to build a down converter that always emits the two photons in the directions I've drawn in the picture, but I don't think it's impossible in principle. Imagine that we have carefully adjusted the lengths of the paths in this experiment so that the lower right detector never clicks, because of destructive interference. All the photons that reach the half-silvered mirror below the detectors will go to the left detector. If we imagine the same setup, but without the half-silvered mirror at the upper right, there wouldn't be any interference at the lower detectors. They would both detect the same number of photons. This is because the upper detectors would give us "which-way information" about the original photon. For example, the detector at the top of the picture will only click if the original photon was reflected at the first half-silvered mirror. The purpose of the half-silvered mirror at the upper right (and the mirrors at the upper left and lower right), is to eliminate the possibility to obtain which-way information that would destroy the interference. There's no way, even in principle, to determine if a photon that's been detected by one of the detectors at the upper right went through the last half-silvered mirror or was reflected by it. Let's move on to the paradox. Suppose that we sometimes remove the half-silvered mirror at the upper right, after one of the lower detectors has clicked. This will make it possible to determine which way the original photon went at the first half-silvered mirror, and this should destroy the interference at the lower two detectors. If the interference really is destroyed, then the lower right detector should detect a photon 50% of the times we do this (instead of 0%). This is retro-causation. The effect precedes the cause. The detector will click because someone at a later time removed a half-silvered mirror. If that isn't strange enough, suppose that we immediately after the lower right detector has detected a photon decide not to remove the half-silvered mirror. Then we have a paradox. If we never remove the half-silvered mirror, that detector should never click. Maybe it's just impossible to destroy the interference by removing the half-silvered mirror after one of the lower detectors has already clicked. But in that case, we can obtain which-way information about the original photon and still get interference. Doesn't this violate the principles of quantum mechanics?