Fixed Points as the Organizing Principle of Functional Iteration

1. Motivation

The equation \(f(f(x)) = F(x)\) looks deceptively simple. One might expect it to behave like an algebraic square root: perhaps one merely "takes half an iterate" in the same way one takes a square root of a number. The difficulty is that composition is geometric rather than multiplicative. The map \(f\) must send points along the orbit structure of \(F\), preserve local regularity, and remain compatible with the fixed-point geometry of the target.

This is why fixed points appear immediately in every serious discussion of half-iterates. If \(x^*\) satisfies \(F(x^*) = x^*\), then the orbit neither escapes nor oscillates at that point; it is the natural place where local linearization, local series methods, and conjugacy-based tools can be anchored. Without such anchors, real half-iterates are often obstructed or become purely local, branch-dependent objects.

Visual guide. The lightweight solver graphs \(F\), \(f\), and \(f(f(x))\) directly, while the accompanying source in resources/manim_half_iterates.py provides a reproducible animation scaffold for orbit and cobweb scenes. Fixed-point and orbit diagram for functional iteration.

Fixed-point and orbit diagram for functional iteration.

2. Fixed Points and Half-Iterates

Let \(F : D \to D\) be a self-map. A functional square root or half-iterate is a map \(f : D \to D\) satisfying \[ f^{\circ 2}(x) = f(f(x)) = F(x). \] A point \(x^*\in D\) is a fixed point of \(F\) if \(F(x^*) = x^*\).

Observation. If \(f\) is a half-iterate of \(F\) and \(f(x^*) = x^*\), then automatically \(F(x^*) = f(f(x^*)) = x^*\). Thus the fixed-point set of \(f\) is contained in the fixed-point set of \(F\), while 2-cycles of \(f\) project to fixed points of \(F\).

The local derivative adds another layer. If \(x^*\) is a smooth fixed point and \(f\) is differentiable, then \[ F'(x^*) = \bigl(f'(x^*)\bigr)^2. \] Hence the multiplier of \(F\) at a fixed point constrains the possible real branches for a local half-iterate. This simple identity already explains why positive, negative, neutral, and zero multipliers behave so differently.

Necessary local condition. A real differentiable half-iterate at a real fixed point requires a real square root of the multiplier \(F'(x^*)\). If \(F'(x^*) < 0\), then a real single-valued local branch is no longer available without leaving the real setting.

3. The Solvable and Unsolvable Contrast

The site's theory page already contrasts \(F(x)=x^2-2\) with \(F(x)=x^2+1\). This contrast is not a curiosity; it is the first dynamical diagnostic one should run before attempting symbolic or numerical construction.

\(x^2-2\)

The graph intersects \(y=x\) at \(-1\) and \(2\), giving real anchors for local study.

\(x^2+1\)

Over the reals the graph stays above \(y=x\), so there is no real fixed point to linearize around.

In the first case, the real fixed points make local conjugacy plausible. In the second case, every real iterate is pushed upward, so one does not even obtain a natural real fixed-point chart. This does not prove that all half-iterates are impossible in other categories, but it explains why the real, explanatory, and browser-friendly approaches in this repository begin from fixed points instead of from global guesswork.

4. Local Conjugacy and Why It Matters

Once a fixed point \(x^*\) is found, set \(y = x - x^*\) and write the shifted map as \[ G(y) = F(x^* + y) - x^*. \] If \(0 < \lambda = G'(0) \neq 1\), then the classical Schröder framework seeks a change of coordinates \(\Phi\) satisfying \[ \Phi(G(y)) = \lambda \, \Phi(y). \] In that coordinate system, iteration becomes multiplication, so the local half-iterate is formally \[ h(y) = \Phi^{-1}\bigl(\sqrt{\lambda}\,\Phi(y)\bigr), \qquad f(x) = x^* + h(x-x^*). \]

This is the logic now exposed in the new analytic solver mode. It does not claim global exactness; instead, it computes a local model, warns when the multiplier is parabolic or non-positive, and verifies the result by checking numerically that \(f(f(x))\) agrees with \(F(x)\) near the chosen fixed point.

Why the warnings matter. Superattracting and parabolic fixed points are not "just another branch" of the same formula. They require different coordinates, such as Böttcher- or Abel-type objects, and those cases are better explained honestly than hidden behind unstable numerics.

5. Applications to the Site Solver

The upgraded solver combines this fixed-point viewpoint with the composita recurrence already developed in the rest of the project.

Paper mode remains the right place for formal power-series roots of \(A^{\circ 2^m}(x)=F(x)\).
Analytic mode begins from a user-chosen fixed point and shows whether the local half-iterate problem is structurally well posed.
Orbit exploration lets the reader see immediately whether a map is attracted, repelled, or pushed away from fixed points.
Graphical verification keeps the project explanatory: formulas are displayed, then checked by recomposition.

For the formal generating-function background, see the composita paper trail collected on the bibliography page; for the broader theory of iterative functional equations, consult the additional references now listed there as well.