
Don’t square? Why not? Okay, fine, we’ll do something else.
The big expression in the question has an obvious recursive structure. If we define , the question’s equation becomes .
Generally speaking, the solutions to an equation of the form are known as the fixed points of . Clearly, the fixed points of any will also be fixed points of This follows immediately from the definition of a fixed point. We can exploit this observation to simplify the problem of finding fixed points of to finding the fixed points of . So let’s find the fixed points of by simply writing the equation and solving it:
With a little massaging, this becomes the cubic
By inspection, is a factor. Dividing by this monomial, we’re left with . After using the quadratic formula, we end up with 3 roots:
These are solutions to the original question, which I confirmed with Octave1.
Three solutions are probably already two more than this question deserves, but I feel compelled to admit that I glossed over something when I mentioned inheriting the fixed points of . It is possible, in general, that has additional fixed points that does not. This is pretty easy to see. Imagine some sends to and to . In that case, and are not fixed points of , but they will be fixed points of . That is, and implies and . The same kind of situation applies to higher-order iterations too.
There is also a graph-theoretic way to view this. Associate with a graph whose vertex set is the union of the domain and range of , and which has an edge wherever . Then 2-cycles in will correspond with loops in . Specifically, every vertex involved in a 2-cycle in will have a loop on it in . In terms of ‘s adjacency matrix , the statement “ may have fixed points that does not” is the same as “ may have nonzero diagonal entries where does not”.
Do any such additional fixed points exist for this particular problem’s ? This addendum has already dragged on, so I leave it as an exercise.
