The evolution of discrete dynamical systems is governed by the iteration of maps. Given an initial condition $x_0$ and a map $f$, the orbit $x_0, f(x_0), f^2(x_0), \ldots$ traces out the long-time behaviour of the system. Fixed points — solutions to $f(p) = p$ — are the simplest invariant objects, and their stability determines whether nearby orbits are attracted to $p$ or driven away from it.

When the derivative $f'(p)$ has modulus strictly different from one, the fixed point is *hyperbolic*, and its stability is resolved entirely by the sign of $|f'(p)| - 1$. Bifurcation theory addresses the complementary case: the *nonhyperbolic* boundary $|f'(p)| = 1$, where the linear approximation is neutral and stability depends on higher-order [derivatives](/page/Derivative). In one dimension, the constraint $f'(p) \in \mathbb{R}$ restricts the critical multiplier to exactly two values — $f'(p) = +1$ and $f'(p) = -1$ — each producing a qualitatively distinct bifurcation. The first leads to the *saddle-node* (tangent) bifurcation, in which two fixed points collide and annihilate. The second leads to the *period-doubling* (flip) bifurcation, in which a fixed point loses stability and a period-2 orbit is born.

This page develops the stability theory at both critical multipliers, introduces the Schwarzian derivative as the natural higher-order invariant controlling the period-doubling case, and classifies the generic saddle-node and period-doubling bifurcations for one-parameter families of maps.

[motivation] ### Linear Stability: What It Achieves The simplest stability criterion for a fixed point $p$ of a smooth map $f$ is *linear*: compute the multiplier $\lambda = f'(p)$ and check whether $|\lambda| < 1$ (stable) or $|\lambda| > 1$ (unstable). The logic is that near $p$, the map is well-approximated by its linearisation $x \mapsto p + \lambda(x - p)$, which contracts distances when $|\lambda| < 1$ and expands them when $|\lambda| > 1$. For the vast majority of fixed points — those with $|\lambda| \neq 1$ — this is the complete story, and no further analysis is needed. ### The Failure at the Critical [Boundary](/page/Boundary) At $|\lambda| = 1$, the linear approximation preserves distances: $|f(x) - p| \approx |x - p|$. The linearisation is neutral, and it cannot distinguish between attraction, repulsion, or any intermediate behaviour. Yet the actual nonlinear map $f$ need not be neutral — the higher-order terms that the linearisation discards are precisely the terms that decide the outcome. The two concrete cases are $\lambda = +1$ and $\lambda = -1$. When $\lambda = +1$, the linearisation is the identity $x \mapsto x$, so the displacement $f(x) - x$ is entirely governed by the nonlinear remainder. When $\lambda = -1$, the linearisation flips each orbit point to the opposite side of $p$, creating an oscillation $x_0, f(x_0), x_2 \approx x_0, \ldots$ whose amplitude is again controlled by the nonlinear terms. ### Two Maps, Same Multiplier, Different Stability To see that higher-order information is essential, consider two maps with the same critical multiplier $\lambda = +1$ at $p = 0$. The map $f(x) = x + x^2$ satisfies $f'(0) = 1$ and $f''(0) = 2 > 0$. Near $x = 0$, the displacement $f(x) - x = x^2 > 0$ for all $x \neq 0$, so every orbit moves to the right: points to the left of $0$ are pushed toward $0$, while points to the right are pushed away. The fixed point is *semi-stable* — attracting from one side and repelling from the other. Now consider $g(x) = x - x^3$, which also satisfies $g'(0) = 1$ but has $g''(0) = 0$ and $g'''(0) = -6 < 0$. Here $g(x) - x = -x^3$, which is positive for $x < 0$ and negative for $x > 0$. Points on both sides are pushed toward $0$, and the fixed point is *asymptotically stable*. The same multiplier produces qualitatively opposite outcomes. The resolution lies in the first nonvanishing higher-order derivative: the sign of $f''(p)$ when it is nonzero, and the sign of $f'''(p)$ when $f''(p) = 0$. Developing this idea rigorously — and extending it to the oscillatory case $\lambda = -1$ via the Schwarzian derivative — is the purpose of this page. [/motivation]

## Definition

The basic objects of study are smooth self-maps of intervals and their fixed points. The regularity requirement $r \ge 3$ ensures that enough derivatives are available for the higher-order stability analysis that follows.

[definition: One-Dimensional Map] Let $I \subseteq \mathbb{R}$ be an interval. A **one-dimensional map** is a [function](/page/Function) \begin{align*} f: I &\to I \end{align*} of class $C^r$ with $r \ge 3$. [/definition]

A map $f$ generates a discrete dynamical system by iteration: starting from $x_0 \in I$, the orbit is the [sequence](/page/Sequence) $x_0, x_1 = f(x_0), x_2 = f(x_1), \ldots$, with $x_n = f^n(x_0)$. The simplest invariant objects for such a system are points that return to themselves after one application of $f$.

[definition: Fixed Point] Let $f \in C^r(I, I)$. A point $p \in I$ is a **fixed point** of $f$ if $f(p) = p$. [/definition]

The existence of a fixed point is an algebraic question — one solves $f(x) = x$. But whether a fixed point is *observable* in practice depends on stability: does a small perturbation from $p$ decay back to $p$ or grow away from it? The derivative of $f$ at $p$ provides the first-order measure of this.

[definition: Multiplier] Let $p$ be a fixed point of $f \in C^r(I, I)$. The **multiplier** of $p$ is the quantity $\lambda := f'(p)$. The fixed point is **hyperbolic** if $|\lambda| \neq 1$ and **nonhyperbolic** if $|\lambda| = 1$. [/definition]

The multiplier governs the local dynamics to first order: near $p$, $f(x) \approx p + \lambda(x - p)$, so the distance $|x - p|$ is scaled by $|\lambda|$ at each iterate. Whether this scaling produces convergence to $p$ is formalised by the following notion.

[definition: Asymptotic Stability Of A Fixed Point] A fixed point $p$ of $f$ is **asymptotically stable** if there exists $\epsilon > 0$ such that for every $x_0 \in (p - \epsilon, p + \epsilon) \cap I$, the orbit satisfies $f^n(x_0) \to p$ as $n \to \infty$. The fixed point is **unstable** if no such neighbourhood exists — equivalently, if every neighbourhood of $p$ contains a point whose orbit eventually leaves that neighbourhood. [/definition]

Between these two extremes lies a third possibility that arises at the critical multiplier $\lambda = +1$: *semi-stability*, where orbits converge from one side but diverge from the other. Semi-stable fixed points are topologically unstable (they do not satisfy the definition of asymptotic stability), but they are significant because they mark the boundary at which two fixed points collide in a saddle-node bifurcation.

## Linear Stability

The first fundamental result confirms the intuition that the multiplier determines stability in the hyperbolic case. When $|\lambda| < 1$, the contraction of the linearisation persists in the full nonlinear map, and when $|\lambda| > 1$, the expansion drives orbits away. The mechanism is elementary — the [Mean Value Theorem](/theorems/186) converts a derivative bound into a geometric contraction or expansion rate — but the result is the foundation on which all subsequent bifurcation analysis rests.

[quotetheorem:641]

The theorem leaves the boundary case $|\lambda| = 1$ completely open. This is not a deficiency of the proof technique but a genuine indeterminacy: as the motivation examples $f(x) = x + x^2$ and $g(x) = x - x^3$ demonstrate, the same critical multiplier can produce semi-stable, stable, or unstable behaviour depending on higher-order terms. The theorem also says nothing about the *rate* of convergence in the stable case, but the proof gives a quantitative bound: the convergence is at least geometric with rate $k$ for any $k \in (|\lambda|, 1)$.

[example: Stability In The Logistic Family] The logistic family \begin{align*} f_\mu: [0, 1] &\to [0, 1] \\ x &\mapsto \mu x(1 - x) \end{align*} with $\mu \in (0, 4]$ provides the simplest illustration. For $\mu > 1$, the nonzero fixed point is \begin{align*} x^* = 1 - \frac{1}{\mu}, \end{align*} obtained by solving $\mu x(1 - x) = x$ and discarding the trivial solution $x = 0$. The multiplier at $x^*$ is \begin{align*} \lambda = f_\mu'(x^*) = \mu(1 - 2x^*) = \mu\left(1 - 2 + \frac{2}{\mu}\right) = 2 - \mu. \end{align*} The stability condition $|\lambda| < 1$ gives $|2 - \mu| < 1$, which holds for $1 < \mu < 3$. At $\mu = 2$, the multiplier is $\lambda = 0$ (superstable); at $\mu = 1$, $\lambda = +1$ (saddle-node); at $\mu = 3$, $\lambda = -1$ (period-doubling). The logistic family thus exhibits both critical multipliers as $\mu$ varies, making it a natural testing ground for the bifurcation theory developed below. [/example]

## The Saddle-Node Bifurcation