The study of discrete dynamical systems begins with the simplest invariant objects: fixed points. A fixed point $p$ of a map $f$ is a state that the system preserves under iteration — once at $p$, the system stays there forever. The fundamental questions are whether the fixed point is stable (do nearby orbits converge to it?) and what happens when stability is lost. For **hyperbolic** fixed points — those whose linearisation has no eigenvalues on the unit circle — the stability picture is completely determined by the eigenvalues: all inside the unit disc means stable, any outside means unstable. The interesting and subtle phenomena occur at the **nonhyperbolic [boundary](/page/Boundary)** $|\lambda| = 1$, where the linear approximation is neutral. Without parameters, the stability at this boundary depends on higher-order [derivatives](/page/Derivative) of the map. With parameters, the passage of eigenvalues through this boundary produces **bifurcations**: qualitative changes in the dynamics such as the creation or destruction of fixed points, the birth of periodic orbits, or the emergence of invariant curves. [motivation] ### What the Linear Approximation Achieves Given a fixed point $p$ of a map $f: \mathbb{R}^n \to \mathbb{R}^n$, the linearisation $x \mapsto Jf_p \cdot x$ determines the local dynamics whenever all eigenvalues $\lambda_i$ satisfy $|\lambda_i| \neq 1$. If every $|\lambda_i| < 1$, each iterate contracts distances and the fixed point is asymptotically stable. If any $|\lambda_i| > 1$, that direction expands and the fixed point is unstable. This is the discrete analogue of the principle that eigenvalues with negative real part give stability for ODEs — but with the unit circle replacing the imaginary axis as the stability boundary. ### Where the Linear Approximation Fails When one or more eigenvalues lie exactly on the unit circle ($|\lambda| = 1$), the linearisation is neutral in the corresponding direction: it neither contracts nor expands. The full nonlinear map may be stable, unstable, or semi-stable depending on the higher-order terms. This is precisely the situation where bifurcation theory becomes necessary: small changes in a parameter can push eigenvalues across the unit circle, switching the fixed point between stable and unstable. The type of bifurcation that occurs depends on *how* the eigenvalues reach the unit circle — whether as a real eigenvalue hitting $+1$, a real eigenvalue hitting $-1$, or a complex conjugate pair crossing the circle. [/motivation] ## Definition [definition: Fixed Point Of A Map] Let $U \subseteq \mathbb{R}^n$ be open and let $f: U \to \mathbb{R}^n$ be a $C^1$ map. A point $p \in U$ is a **fixed point** of $f$ if $f(p) = p$. [/definition] The local behaviour near a fixed point is governed by the linearisation. For a one-dimensional map $f: I \to I$ with $I \subseteq \mathbb{R}$, the relevant quantity is the single derivative $f'(p)$. For maps on $\mathbb{R}^n$ with $n \ge 2$, the linearisation is encoded in the Jacobian matrix. [definition: Multipliers Of A Fixed Point] Let $f: U \subseteq \mathbb{R}^n \to \mathbb{R}^n$ be a $C^1$ map with a fixed point $p \in U$. The **Jacobian matrix** of $f$ at $p$ is \begin{align*} Jf_p \in \mathbb{R}^{n \times n}, \qquad (Jf_p)_{ij} = \frac{\partial f_i}{\partial x_j}(p). \end{align*} The eigenvalues $\lambda_1, \ldots, \lambda_n$ of $Jf_p$ (counted with algebraic multiplicity, possibly complex) are the **multipliers** of the fixed point. [/definition] Since $Jf_p$ is a real matrix, complex eigenvalues come in conjugate pairs $\lambda, \bar{\lambda}$, which necessarily share the same modulus. The multipliers determine whether the linearisation contracts or expands near the fixed point, and their position relative to the unit circle governs stability. [definition: Hyperbolic Fixed Point Of A Map] A fixed point $p$ of a $C^1$ map $f: U \subseteq \mathbb{R}^n \to \mathbb{R}^n$ is **hyperbolic** if every multiplier satisfies $|\lambda_i| \neq 1$. If all $|\lambda_i| < 1$, the fixed point is a **sink**. If all $|\lambda_i| > 1$, it is a **source**. If some multipliers are inside and some outside the unit disc, it is a **saddle**. A fixed point is **nonhyperbolic** if at least one multiplier satisfies $|\lambda_i| = 1$. [/definition] ## Stability of Fixed Points ### The Hyperbolic Case When all multipliers lie strictly inside the unit disc, the linearisation is a uniform contraction: there exist constants $C > 0$ and $\rho \in (0,1)$ such that $|f^k(x) - p| \le C \rho^k |x - p|$ for all $x$ sufficiently close to $p$ and all $k \ge 0$. The fixed point attracts all nearby orbits at a geometric rate. When any multiplier lies outside the unit disc, the corresponding eigendirection expands under iteration, and orbits are repelled from $p$. [theorem: Linear Stability Of Fixed Points Of Maps] Let $f: U \subseteq \mathbb{R}^n \to \mathbb{R}^n$ be a $C^1$ map with a fixed point $p \in U$, and let $\lambda_1, \ldots, \lambda_n$ be the multipliers of $p$. 1. If $|\lambda_i| < 1$ for all $i = 1, \ldots, n$, then $p$ is **asymptotically stable**: there exists a neighbourhood $V$ of $p$ such that $f^k(x) \to p$ as $k \to \infty$ for every $x \in V$. 2. If $|\lambda_i| > 1$ for some $i$, then $p$ is **unstable**: for every neighbourhood $V$ of $p$ and every $x \in V$ with $x \neq p$, the orbit $(f^k(x))_{k \ge 0}$ eventually leaves $V$. [/theorem] In one dimension, this reduces to the [Linear Stability of Fixed Points of One-Dimensional Maps](/theorems/641): the single multiplier $\lambda = f'(p)$ determines stability via $|\lambda| < 1$ (stable) or $|\lambda| > 1$ (unstable). The theorem makes no claim about the case $|\lambda_i| = 1$ for some $i$ — this is precisely the nonhyperbolic boundary where the linear approximation is inconclusive. ### The Nonhyperbolic Boundary When one or more multipliers lie on the unit circle, the linearisation is neutral in the corresponding directions and the stability of the fixed point depends on the higher-order terms of $f$. Two maps with the same linearisation can exhibit opposite stability behaviour: for instance, $f(x) = x + x^2$ and $g(x) = x - x^3$ both have $f'(0) = g'(0) = 1$, but the origin is semi-stable for $f$ (attracting from one side only) and asymptotically stable for $g$. In one dimension, the nonhyperbolic boundary consists of exactly two values: $\lambda = +1$ and $\lambda = -1$. The stability at each is resolved by specific higher-order derivatives: At $\lambda = +1$, the [Stability of a Nonhyperbolic Fixed Point with Multiplier One](/theorems/642) shows that if $f''(p) \neq 0$, the fixed point is **semi-stable** (attracting from one side, repelling from the other); if $f''(p) = 0$, the sign of $f'''(p)$ determines full stability or instability. At $\lambda = -1$, the [Schwarzian Derivative Stability Criterion](/theorems/643) shows that the sign of $Sf(p) = f'''(p)/f'(p) - \frac{3}{2}(f''(p)/f'(p))^2$ determines stability: $Sf(p) < 0$ gives stability, $Sf(p) > 0$ gives instability. In higher dimensions, the [Center Manifold Theorem](/page/Center%20Manifold%20Theorem) reduces the nonhyperbolic analysis to the center manifold — the invariant manifold tangent to the eigenspace of the unit-circle eigenvalues. The restricted dynamics on this lower-dimensional manifold then falls into one of the cases above. ### Why This Motivates Bifurcation Theory The stability analysis above applies to a **single, fixed map**. It tells us whether a given nonhyperbolic fixed point is stable or not by examining higher-order derivatives. But there is a more fundamental question: if we have a **family** of maps $x \mapsto f(x, \alpha)$ depending on a parameter $\alpha$, and the multipliers cross the unit circle as $\alpha$ varies, what new dynamical objects are created or destroyed at the crossing? This is the domain of bifurcation theory, which we develop next. ## Bifurcations in Families of Maps Consider a smooth family of maps depending on parameters: \begin{align*} x \mapsto f(x, \alpha), \quad x \in \mathbb{R}^n, \quad \alpha \in \mathbb{R}^m, \end{align*} where $f$ is at least $C^3$ and $f(p(\alpha), \alpha) = p(\alpha)$ traces a smooth family of fixed points. The multipliers $\lambda_i(\alpha)$ of the Jacobian $Jf_{p(\alpha)}$ now depend on the parameter. A **bifurcation** occurs at a parameter value $\alpha^*$ where the qualitative dynamics change — typically because one or more multipliers cross the unit circle. ### The Three Cases at the Unit Circle The eigenvalues of the real matrix $Jf_p$ are either real or come in complex conjugate pairs. When a multiplier reaches the unit circle $|\lambda| = 1$, exactly three mutually exclusive scenarios arise: 1. **A real eigenvalue reaches $+1$.** This happens when $\det(Jf_p - I) = 0$. The corresponding bifurcation is the **saddle-node** (fold): two fixed points collide and annihilate, or a single fixed point splits into two. 2. **A real eigenvalue reaches $-1$.** This happens when $\det(Jf_p + I) = 0$. The corresponding bifurcation is **period-doubling** (flip): the fixed point loses stability and a period-2 orbit is born. 3. **A complex conjugate pair reaches the unit circle.** This happens when $\lambda, \bar{\lambda} = e^{\pm i\theta}$ with $\theta \in (0, \pi)$, so both eigenvalues cross the circle simultaneously (they share the same modulus). The corresponding bifurcation is the **Neimark-Sacker** bifurcation: an invariant closed curve branches from the fixed point. These three cases exhaust all possibilities: a real number on the unit circle is either $+1$ or $-1$, and non-real eigenvalues of a real matrix are conjugate pairs with equal modulus. [remark: Level Of Analysis] Locating bifurcation curves (finding the parameter values where eigenvalues hit the unit circle) requires only the **necessary eigenvalue conditions** listed above. Determining what exactly happens at the bifurcation — whether it is supercritical or subcritical, the stability of the bifurcating objects, and the direction in parameter space — requires the full bifurcation theorems with their non-degeneracy and transversality conditions, including Lyapunov coefficients. [/remark] ### Saddle-Node Bifurcation When a single real multiplier crosses $+1$, the graph of $f$ becomes tangent to the identity at the fixed point. Geometrically, two fixed points that existed on opposite sides of a tangency collide and disappear — or, in reverse, a degenerate fixed point splits into two (one stable, one unstable). [definition: Saddle-Node Bifurcation Condition For Maps] Let $f: \mathbb{R}^n \times \mathbb{R}^m \to \mathbb{R}^n$ be a smooth family of maps with a fixed point $p$ at parameter value $\alpha^*$. The fixed point satisfies the **necessary condition for a saddle-node bifurcation** if $+1$ is an eigenvalue of $Jf_p$: \begin{align*} \det(Jf_p - I) = 0. \end{align*} For the bifurcation to be generic, the remaining multipliers should satisfy $|\lambda_i| \neq 1$, and appropriate non-degeneracy and transversality conditions must hold. The full one-dimensional theory is developed in [Bifurcation Theory of One-Dimensional Maps](/page/Bifurcation%20Theory%20of%20One-Dimensional%20Maps). [/definition] The condition $\det(Jf_p - I) = 0$ and the fixed-point equation $f(p, \alpha) = p$ together define a codimension-1 surface in parameter space — the **saddle-node bifurcation curve**. ### Period-Doubling Bifurcation When a single real multiplier crosses $-1$, the linearisation reverses direction each iterate: $x, f(x), f^2(x), \ldots$ alternates about the fixed point. Since the second iterate $f^2$ has multiplier $(-1)^2 = 1$ at the fixed point, this places $p$ at the saddle-node boundary for $f^2$. The result is that a pair of fixed points of $f^2$ branches off — which, viewed as orbits of $f$, form a **period-2 orbit**. [definition: Period-Doubling Bifurcation Condition For Maps] Let $f: \mathbb{R}^n \times \mathbb{R}^m \to \mathbb{R}^n$ be a smooth family of maps with a fixed point $p$ at parameter value $\alpha^*$. The fixed point satisfies the **necessary condition for a period-doubling bifurcation** if $-1$ is an eigenvalue of $Jf_p$: \begin{align*} \det(Jf_p + I) = 0. \end{align*} For the bifurcation to be generic, the remaining multipliers should satisfy $|\lambda_i| \neq 1$. In one dimension, the supercritical/subcritical character is determined by the sign of the [Schwarzian Derivative Stability Criterion](/theorems/643): $Sf(p) < 0$ gives a stable period-2 orbit (supercritical), $Sf(p) > 0$ gives an unstable one (subcritical). [/definition] ### Neimark-Sacker Bifurcation When a complex conjugate pair $\lambda(\alpha), \overline{\lambda(\alpha)} = r(\alpha) e^{\pm i\theta(\alpha)}$ crosses the unit circle, the fixed point transitions from a stable spiral (eigenvalues inside the unit disc, $r < 1$) to an unstable spiral (eigenvalues outside, $r > 1$). At the moment of crossing, the linearisation is an exact rotation by angle $\theta_0$, and the nonlinear terms determine whether an invariant closed curve branches off from the fixed point. For a two-dimensional map, the eigenvalue product equals the determinant: $\lambda \bar{\lambda} = r^2 = \det(Jf_p)$. The condition $r = 1$ is therefore equivalent to $\det(Jf_p) = 1$, and the eigenvalues are complex (not real) when $(\operatorname{tr} Jf_p)^2 < 4 \det(Jf_p)$, i.e., $|\operatorname{tr} Jf_p| < 2$. The full theorem requires non-resonance ($e^{ik\theta_0} \neq 1$ for $k = 1, 2, 3, 4$, equivalently $\theta_0 \notin \{0, \pi/2, 2\pi/3, \pi\}$), transversality ($r'(0) \neq 0$), and non-degeneracy of the First Lyapunov Coefficient $L_1$. When these conditions hold, a unique invariant closed curve bifurcates from the fixed point. [quotetheorem:235] The sign of $L_1$ determines the nature of the bifurcation: $L_1 < 0$ (supercritical) produces a stable invariant curve that attracts nearby orbits, while $L_1 > 0$ (subcritical) produces an unstable invariant curve that repels them. The dynamics on the invariant curve can be quasi-periodic (orbits dense on the curve, when the rotation number $\theta_0/(2\pi)$ is irrational) or phase-locked (finitely many periodic orbits, when the rotation number is rational). A detailed treatment including worked computations of $L_1$ is given in [Neimark-Sacker Bifurcation](/page/Neimark-Sacker%20Bifurcation). ## Computing Bifurcation Curves In many exam and applied problems, the first task is to locate the bifurcation curves in parameter space — the parameter values at which each type of bifurcation can occur. This requires only the necessary eigenvalue conditions developed above. ### Recipe for Two-Dimensional Maps For a two-dimensional map $x \mapsto f(x, \alpha)$ with parameters $\alpha \in \mathbb{R}^m$: **Step 1.** Find all fixed points by solving $f(p, \alpha) = p$. Express the fixed point coordinates in terms of the parameters. **Step 2.** Compute the Jacobian $J = Jf_p$ at each fixed point. The characteristic equation for a $2 \times 2$ matrix is $\lambda^2 - (\operatorname{tr} J)\lambda + \det J = 0$. **Step 3.** Apply each eigenvalue condition: - **Saddle-node:** $\det(J - I) = 0$. This yields a relation between fixed point coordinates and parameters. - **Period-doubling:** $\det(J + I) = 0$. Same type of relation. - **Neimark-Sacker:** $\det J = 1$ (eigenvalues on the unit circle) **and** $(\operatorname{tr} J)^2 < 4$ (eigenvalues are complex, i.e. $|\operatorname{tr} J| < 2$). **Step 4.** Eliminate the fixed point coordinates using the fixed-point equation from Step 1 to obtain curves (or surfaces) in parameter space. [example: Bifurcation Curves Of A Two-Parameter Planar Map] Consider the map \begin{align*} \bar{x} &= y, \\ \bar{y} &= a - bx - y^2, \end{align*} with parameters $a$ and $b$. **Step 1: Fixed points.** Setting $\bar{x} = x$ and $\bar{y} = y$ gives $x = y$ and $y = a - by - y^2$. Substituting: \begin{align*} y^2 + (1 + b)y - a = 0. \end{align*} This is quadratic in $y$, so there are at most two fixed points, each of the form $(y, y)$. **Step 2: Jacobian.** At a fixed point $(y, y)$: \begin{align*} J = \begin{pmatrix} 0 & 1 \\ -b & -2y \end{pmatrix}, \quad \operatorname{tr} J = -2y, \quad \det J = b. \end{align*} The characteristic equation is $\lambda^2 + 2y\lambda + b = 0$, with eigenvalue product $\lambda_1 \lambda_2 = b$ and eigenvalue sum $\lambda_1 + \lambda_2 = -2y$. **Step 3a: Saddle-node ($\lambda = +1$).** Computing $\det(J - I) = 0$: \begin{align*} \det\begin{pmatrix} -1 & 1 \\ -b & -2y - 1 \end{pmatrix} = (-1)(-2y-1) - (1)(-b) = 2y + 1 + b = 0, \end{align*} giving $y = -(1+b)/2$. Substituting into the fixed-point equation: \begin{align*} \frac{(1+b)^2}{4} + (1+b) \cdot \frac{-(1+b)}{2} - a &= 0 \\ \frac{(1+b)^2}{4} - \frac{(1+b)^2}{2} - a &= 0 \\ a &= -\frac{(1+b)^2}{4}. \end{align*} The saddle-node curve is $s_1 = \left\{(a, b) : a = -\dfrac{(1+b)^2}{4}\right\}$. **Consistency check.** When $\lambda_1 = 1$, the other eigenvalue is $\lambda_2 = b$ (from the product $\lambda_1 \lambda_2 = b$). This is different from $+1$ as long as $b \neq 1$, confirming that only one eigenvalue is at the critical value generically. **Step 3b: Period-doubling ($\lambda = -1$).** Computing $\det(J + I) = 0$: \begin{align*} \det\begin{pmatrix} 1 & 1 \\ -b & -2y + 1 \end{pmatrix} = (1)(-2y+1) - (1)(-b) = -2y + 1 + b = 0, \end{align*} giving $y = (1+b)/2$. Substituting into the fixed-point equation: \begin{align*} \frac{(1+b)^2}{4} + (1+b) \cdot \frac{(1+b)}{2} - a &= 0 \\ a &= \frac{3(1+b)^2}{4}. \end{align*} The period-doubling curve is $s_2 = \left\{(a, b) : a = \dfrac{3(1+b)^2}{4}\right\}$. **Step 3c: Neimark-Sacker (complex pair on unit circle).** The eigenvalue product equals $\det J = b$. For a conjugate pair $\lambda, \bar{\lambda} = re^{\pm i\theta}$, we have $r^2 = b$, so $|\lambda| = 1$ requires $b = 1$. The eigenvalues are complex when $(\operatorname{tr} J)^2 < 4\det J$, i.e., $4y^2 < 4$, giving $|y| < 1$. From the fixed-point equation with $b = 1$: $y^2 + 2y - a = 0$, so $y = -1 \pm \sqrt{1 + a}$. Applying $|y| < 1$: For $y = -1 + \sqrt{1+a}$: the inequality $|{-1 + \sqrt{1+a}}| < 1$ gives $0 < \sqrt{1+a} < 2$, hence $-1 < a < 3$. For $y = -1 - \sqrt{1+a}$: the inequality $|{-1 - \sqrt{1+a}}| < 1$ requires $\sqrt{1+a} < 0$, which is impossible. The Neimark-Sacker curve is $s_3 = \{(a, b) : -1 < a < 3,\; b = 1\}$. **Intersection points.** The saddle-node curve $s_1$ meets $b = 1$ at $a = -1$ (where $y = -1$, so $\lambda = e^{\pm i\pi} = -1$, a double eigenvalue at $-1$). The period-doubling curve $s_2$ meets $b = 1$ at $a = 3$ (where $y = 1$, also giving a special eigenvalue configuration). These intersection points are codimension-2 bifurcation points where two bifurcation types interact. [/example] ## References 1. Y. A. Kuznetsov, *Elements of Applied Bifurcation Theory*, 3rd ed., Springer (2004). 2. J. Guckenheimer and P. Holmes, *Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields*, Springer (1983). 3. S. Wiggins, *Introduction to Applied Nonlinear Dynamical Systems and Chaos*, 2nd ed., Springer (2003). 4. V. I. Arnold, *Geometrical Methods in the Theory of Ordinary Differential Equations*, 2nd ed., Springer (1988).