Fixed Points and Equilibria

Suppose a population of foxes and rabbits evolves according to some system of nonlinear differential equations. Is there a population level at which nothing ever changes again? This question — does a dynamical system have a state of perfect balance — is the entry point to almost all of qualitative Autonomous Systems and Phase Space analysis. Long before we can ask whether such a balance is stable, attracting, or physically observable, we must first know how to find it. We must also know what it means mathematically. Fixed points, also called equilibria or critical points, are the skeleton around which the entire phase portrait of a nonlinear system is organized. Trajectories bend around them, spiral into or away from them, and get funneled between them. Nearly every qualitative feature of a flow can be traced back to the location and character of its equilibria. Locating them is often the very first, and computationally simplest, step in analyzing any autonomous system. Doing so converts a calculus problem (solving differential equations) into an algebra problem (solving a system of equations). The big picture is this: an autonomous system x˙=f(x)\dot{\mathbf{x}} = \mathbf{f}(\mathbf{x}) assigns to every point in phase space a velocity vector. A fixed point is a point where that vector vanishes. There, the flow has, in a precise sense, nowhere to go. This article makes that idea completely rigorous. It proves that fixed points correspond exactly to constant solutions of the system. It also builds the intuition and computational habits needed before tackling Linearization and the Jacobian and Stability Concepts and Definitions. There, we ask what happens near an equilibrium rather than merely where it sits. By the end, you will be able to state the precise definition of a fixed point. You will be able to prove the correspondence between fixed points and constant solutions using the uniqueness theorem for ODEs. You will work through two asymmetric nonlinear examples by hand. You will also recognize the common errors — degenerate systems, non-isolated equilibria, and confusing zeros of f\mathbf{f} with zeros of a single component — that trip up students moving from linear to nonlinear systems.

Key takeaways
  • A fixed point of x˙=f(x)\dot{\mathbf{x}}=\mathbf{f}(\mathbf{x}) is a point x∗\mathbf{x}^* with f(x∗)=0\mathbf{f}(\mathbf{x}^*)=\mathbf{0}.
  • Fixed points correspond exactly to constant (equilibrium) solutions x(t)≡x∗\mathbf{x}(t)\equiv \mathbf{x}^*.
  • Finding them reduces a differential equation problem to an algebraic root-finding problem.
  • In nn dimensions you must solve nn simultaneous equations f1=⋯=fn=0f_1=\cdots=f_n=0 at once, not one at a time.
  • Equilibria organize the phase portrait: trajectories cannot cross them and often approach or flee from them.
  • Key idea: setting the vector field to zero converts "when is the system unchanging?" into solvable algebra, and every such solution is a genuine constant trajectory.

Where this lives today

Fixed-point analysis of vector fields underlies population ecology (coexistence and extinction equilibria in Lotka–Volterra-type models), chemical kinetics (steady states of reaction networks), epidemiology (disease-free and endemic equilibria in SIR-type models), electrical circuit theory (DC operating points of nonlinear circuits), control theory (equilibria that a controller must stabilize), economics (market or general equilibria as zeros of excess-demand vector fields), and computer science (fixed points of iterative and recurrent algorithms, including training dynamics and recurrent neural network steady states). In statistics and machine learning, fixed points of expectation-maximization or gradient-flow updates play an analogous role, and in cryptography, fixed points of certain modular maps appear in the analysis of pseudorandom generators. In every case, the first mathematical move is identical to what is done here: set the governing vector field to zero and solve.

Ask the tutor: How do I linearize a nonlinear system near one of its fixed points to classify its stability?

DefinitionFixed point of an autonomous system

Let D⊆RnD \subseteq \mathbb{R}^n be an open set and let f:D→Rn\mathbf{f}: D \to \mathbb{R}^n be a vector field, giving rise to the autonomous system

x˙=f(x),x(t)∈D.\dot{\mathbf{x}} = \mathbf{f}(\mathbf{x}), \qquad \mathbf{x}(t) \in D.

A point x∗∈D\mathbf{x}^* \in D is called a fixed point (equivalently, an equilibrium point or critical point) of the system if

f(x∗)=0.\mathbf{f}(\mathbf{x}^*) = \mathbf{0}.

The set of all fixed points is

E={x∗∈D:f(x∗)=0}=f−1(0).E = \{\mathbf{x}^* \in D : \mathbf{f}(\mathbf{x}^*) = \mathbf{0}\} = \mathbf{f}^{-1}(\mathbf{0}).

A fixed point x∗\mathbf{x}^* is called isolated if there is an open neighborhood UU of x∗\mathbf{x}^* such that U∩E={x∗}U \cap E = \{\mathbf{x}^*\}.

Intuition

Picture the phase plane of a two-dimensional system as a field of arrows, one arrow f(x,y)=(f1(x,y),f2(x,y))\mathbf{f}(x,y) = (f_1(x,y), f_2(x,y)) attached to every point, indicating the instantaneous direction and speed of motion of a trajectory passing through that point. A fixed point is a location where the arrow has zero length — the wind, so to speak, has died down completely. Any object placed exactly there stays there forever, because there is no velocity to carry it away. This is exactly analogous to critical points of a single autonomous first-order equation x˙=f(x)\dot{x} = f(x), which you may have already met as the roots of ff plotted on a phase line. The step to nn dimensions is conceptually the same — solve f(x)=0\mathbf{f}(\mathbf{x}) = \mathbf{0} — but algebraically harder, because now nn equations in nn unknowns must hold simultaneously. A common beginner's error is solving f1=0f_1 = 0 and f2=0f_2 = 0 separately and pairing up roots that do not actually correspond to the same point; the correct approach is to solve the system jointly, typically by using one equation to eliminate a variable in the other. Geometrically, fixed points are the places where trajectories can "get stuck" or, viewed backward in time, where they can have originated infinitely long ago. Nearby trajectories are shaped by how the vector field behaves in a neighborhood of each equilibrium: they may spiral toward it, spiral away, be pulled along one direction and repelled along another (a saddle-like structure), or circulate around it in closed loops. The classification of these behaviors is the subject of Linearization and the Jacobian; here we only need to locate the equilibria as the first mandatory step.

Figure
Vector field arrows and zeros in the phase plane
x y x* = (−1, −0.5) x** = (1.8, 0.8)
TheoremCorrespondence between fixed points and constant solutions

Let D⊆RnD \subseteq \mathbb{R}^n be open and let f:D→Rn\mathbf{f}: D \to \mathbb{R}^n be continuous. Consider the autonomous system

x˙(t)=f(x(t)).\dot{\mathbf{x}}(t) = \mathbf{f}(\mathbf{x}(t)).

Then x∗∈D\mathbf{x}^* \in D is a fixed point of f\mathbf{f} (i.e. f(x∗)=0\mathbf{f}(\mathbf{x}^*) = \mathbf{0}) if and only if the constant function x(t)≡x∗\mathbf{x}(t) \equiv \mathbf{x}^* for all t∈Rt \in \mathbb{R} is a solution of the system. Moreover, if f\mathbf{f} is additionally assumed to be locally Lipschitz on DD, then for each fixed point x∗\mathbf{x}^* this constant solution is the unique solution of the system with initial condition x(t0)=x∗\mathbf{x}(t_0) = \mathbf{x}^* for any t0t_0; in particular no non-constant trajectory can pass through x∗\mathbf{x}^*.

Proof

We prove the two directions of the "if and only if" separately, then prove the uniqueness statement.

(⇐\Leftarrow) If x(t)≡x∗\mathbf{x}(t) \equiv \mathbf{x}^* solves the system, then f(x∗)=0\mathbf{f}(\mathbf{x}^*) = \mathbf{0}. Suppose x(t)=x∗\mathbf{x}(t) = \mathbf{x}^* for all tt is a differentiable function satisfying x˙(t)=f(x(t))\dot{\mathbf{x}}(t) = \mathbf{f}(\mathbf{x}(t)) on some interval. Since x(t)\mathbf{x}(t) is constant in tt, its derivative is identically zero:

x˙(t)=ddtx∗=0for all t.\dot{\mathbf{x}}(t) = \frac{d}{dt}\mathbf{x}^* = \mathbf{0} \quad \text{for all } t.

Because x(t)\mathbf{x}(t) solves the ODE, we also have x˙(t)=f(x(t))=f(x∗)\dot{\mathbf{x}}(t) = \mathbf{f}(\mathbf{x}(t)) = \mathbf{f}(\mathbf{x}^*) for all tt (using x(t)≡x∗\mathbf{x}(t) \equiv \mathbf{x}^*). Equating the two expressions for x˙(t)\dot{\mathbf{x}}(t) gives

0=f(x∗).\mathbf{0} = \mathbf{f}(\mathbf{x}^*).

Hence x∗\mathbf{x}^* is a fixed point. This direction used only the definition of derivative of a constant function and the definition of "solving the ODE"; no additional hypotheses on f\mathbf{f} beyond it being defined at x∗\mathbf{x}^* were needed.

(⇒\Rightarrow) If f(x∗)=0\mathbf{f}(\mathbf{x}^*) = \mathbf{0}, then x(t)≡x∗\mathbf{x}(t) \equiv \mathbf{x}^* solves the system. Define x:R→D\mathbf{x}: \mathbb{R} \to D by x(t)=x∗\mathbf{x}(t) = \mathbf{x}^* for all tt. This function is differentiable (it is constant), and its derivative is

x˙(t)=0for all t,\dot{\mathbf{x}}(t) = \mathbf{0} \quad \text{for all } t,

directly from the definition of the derivative of a constant map: for any tt,

x˙(t)=lim⁡h→0x(t+h)−x(t)h=lim⁡h→0x∗−x∗h=0.\dot{\mathbf{x}}(t) = \lim_{h \to 0} \frac{\mathbf{x}(t+h) - \mathbf{x}(t)}{h} = \lim_{h\to 0} \frac{\mathbf{x}^* - \mathbf{x}^*}{h} = \mathbf{0}.

On the other hand, since x(t)=x∗\mathbf{x}(t) = \mathbf{x}^* for every tt, continuity of f\mathbf{f} (assumed) lets us evaluate

f(x(t))=f(x∗)=0\mathbf{f}(\mathbf{x}(t)) = \mathbf{f}(\mathbf{x}^*) = \mathbf{0}

for every tt, using the hypothesis f(x∗)=0\mathbf{f}(\mathbf{x}^*) = \mathbf{0}. Comparing the two computations,

x˙(t)=0=f(x(t))for all t,\dot{\mathbf{x}}(t) = \mathbf{0} = \mathbf{f}(\mathbf{x}(t)) \quad \text{for all } t,

so x(t)≡x∗\mathbf{x}(t) \equiv \mathbf{x}^* satisfies the differential equation for all t∈Rt \in \mathbb{R}. This establishes the equivalence.

Uniqueness statement. Now additionally assume f\mathbf{f} is locally Lipschitz on DD: for every point p∈D\mathbf{p} \in D there is a neighborhood UU of p\mathbf{p} and a constant L≥0L \geq 0 such that

∥f(u)−f(v)∥≤L∥u−v∥for all u,v∈U.\|\mathbf{f}(\mathbf{u}) - \mathbf{f}(\mathbf{v})\| \leq L\|\mathbf{u}-\mathbf{v}\| \quad \text{for all } \mathbf{u},\mathbf{v} \in U.

This is exactly the hypothesis of the Picard–Lindelöf existence-and-uniqueness theorem for the initial value problem

x˙=f(x),x(t0)=x0,\dot{\mathbf{x}} = \mathbf{f}(\mathbf{x}), \qquad \mathbf{x}(t_0) = \mathbf{x}_0,

which guarantees, for any t0∈Rt_0 \in \mathbb{R} and x0∈D\mathbf{x}_0 \in D, the existence of ε>0\varepsilon > 0 and a unique solution x(t)\mathbf{x}(t) defined on (t0−ε,t0+ε)(t_0 - \varepsilon, t_0+\varepsilon) satisfying the initial value problem. We invoke this classical theorem as a known prerequisite result — it does not depend on the present theorem, so there is no circularity. Fix a fixed point x∗\mathbf{x}^* and t0∈Rt_0 \in \mathbb{R}, and consider the initial value problem x˙=f(x)\dot{\mathbf{x}} = \mathbf{f}(\mathbf{x}), x(t0)=x∗\mathbf{x}(t_0) = \mathbf{x}^*. By the direction (⇒\Rightarrow) proved above, the constant function x(t)≡x∗\mathbf{x}(t) \equiv \mathbf{x}^* is a solution of this initial value problem, since it solves the ODE for all tt and in particular satisfies x(t0)=x∗\mathbf{x}(t_0) = \mathbf{x}^*. By the uniqueness clause of Picard–Lindelöf, this is the only solution on any interval containing t0t_0 where a solution is defined. Consequently, if any trajectory y(t)\mathbf{y}(t) of the system satisfies y(t1)=x∗\mathbf{y}(t_1) = \mathbf{x}^* for some time t1t_1, then by uniqueness applied at t0=t1t_0 = t_1, y(t)≡x∗\mathbf{y}(t) \equiv \mathbf{x}^* for all tt in its domain of definition. Hence no non-constant trajectory can pass through x∗\mathbf{x}^*, completing the proof. ■\blacksquare

∎

Worked examples

Example 1 (asymmetric planar system). Consider the nonlinear system

x˙=x2−y−2,y˙=x−y2+4.\dot{x} = x^2 - y - 2, \qquad \dot{y} = x - y^2 + 4.

We seek all (x,y)(x,y) with x2−y−2=0x^2 - y - 2 = 0 and x−y2+4=0x - y^2 + 4 = 0 simultaneously. From the first equation, solve for yy in terms of xx:

y=x2−2.y = x^2 - 2.

Substitute into the second equation:

x−(x2−2)2+4=0.x - (x^2-2)^2 + 4 = 0.

Expand (x2−2)2=x4−4x2+4(x^2-2)^2 = x^4 - 4x^2 + 4, so

x−x4+4x2−4+4=0⟹−x4+4x2+x=0⟹x4−4x2−x=0.x - x^4 + 4x^2 - 4 + 4 = 0 \quad\Longrightarrow\quad -x^4 + 4x^2 + x = 0 \quad\Longrightarrow\quad x^4 - 4x^2 - x = 0.

Factor out xx:

x(x3−4x−1)=0.x(x^3 - 4x - 1) = 0.

One root is immediately x=0x = 0. For the cubic x3−4x−1=0x^3 - 4x - 1 = 0, check for rational roots among ±1\pm 1: at x=1x=1, 1−4−1=−4≠01 - 4 - 1 = -4 \neq 0; at x=−1x=-1, −1+4−1=2≠0-1+4-1=2 \neq 0; no rational root, so its three real (or complex) roots are irrational. Evaluate the cubic at a few points to confirm three real roots exist: at x=−2x=-2, (−8)−(−8)−1=−1(-8) - (-8) - 1 = -1; at x=−1.9x=-1.9: (−6.859)−(−7.6)−1=−0.259(-6.859) - (-7.6) - 1 = -0.259; at x=−1.8x = -1.8: (−5.832)−(−7.2)−1=0.368(-5.832)-(-7.2)-1=0.368, so a root lies in (−1.9,−1.8)(-1.9,-1.8). At x=0x=0: −1<0-1 < 0; at x=−0.3x=-0.3: −0.027+1.2−1=0.173>0-0.027+1.2-1=0.173>0, so a root lies in (−0.3,0)(-0.3,0). At x=2x=2: 8−8−1=−1<08-8-1=-1<0; at x=2.2x=2.2: 10.648−8.8−1=0.848>010.648-8.8-1=0.848>0, so a root lies in (2,2.2)(2,2.2). Hence the cubic contributes three real roots, approximately x≈−1.86, −0.25, 2.11x \approx -1.86,\ -0.25,\ 2.11 (refining slightly: at x=−1.86x=-1.86, (−6.435)−(−7.44)−1=0.005(-6.435)-(-7.44)-1=0.005, close enough; at x=−0.25x=-0.25: −0.0156+1−1=−0.0156-0.0156+1-1=-0.0156, so refine to x≈−0.254x\approx -0.254; at x=2.11x=2.11: 9.394−8.44−1=−0.0469.394-8.44-1=-0.046, refine to x≈2.115x \approx 2.115). So altogether there are four fixed points, with y=x2−2y = x^2-2 in each case:

  • x=0x=0: y=−2y = -2. Fixed point (0,−2)(0,-2) — exact.
  • x≈−1.86x \approx -1.86: y≈(−1.86)2−2=3.460−2=1.460y \approx (-1.86)^2 - 2 = 3.460-2 = 1.460.
  • x≈−0.254x \approx -0.254: y≈0.0645−2=−1.935y \approx 0.0645-2=-1.935.
  • x≈2.115x \approx 2.115: y≈4.473−2=2.473y \approx 4.473-2=2.473.

Let us verify the exact fixed point (0,−2)(0,-2) directly in the original equations: x˙=02−(−2)−2=2−2=0\dot x = 0^2-(-2)-2 = 2-2=0 ✓, and y˙=0−(−2)2+4=0−4+4=0\dot y = 0 - (-2)^2+4 = 0-4+4=0 ✓. Both vanish, confirming (0,−2)(0,-2) is indeed a fixed point. By the theorem just proved, the constant function (x(t),y(t))≡(0,−2)(x(t),y(t)) \equiv (0,-2) is an exact solution of this nonlinear system for all time — even though the system itself has no closed-form general solution. This example illustrates the essential technique: use one equation to express one variable in terms of the other, substitute into the second equation to get a single-variable polynomial, and solve that polynomial (exactly when possible, numerically otherwise). It also shows that a nonlinear system typically has multiple isolated fixed points, unlike a generic linear system x˙=Ax\dot{\mathbf{x}} = A\mathbf{x} with det⁡A≠0\det A \neq 0, which has exactly one (the origin).

Example 2 (a population-type model with a curve of interacting nonlinearities). Consider the Lotka–Volterra-flavored system

x˙=x(3−x−2y),y˙=y(2x−1−y),\dot{x} = x(3 - x - 2y), \qquad \dot{y} = y(2x - 1 - y),

representing, say, a species xx and species yy with population-dependent growth rates. We find all fixed points. Fixed points require x(3−x−2y)=0x(3-x-2y) = 0 and y(2x−1−y)=0y(2x-1-y)=0 simultaneously. Each equation is a product, so we case-split. Case A: x=0x = 0. The second equation becomes y(0−1−y)=0y(0 - 1 - y) = 0, i.e. y(−1−y)=0y(-1-y)=0, so y=0y = 0 or y=−1y=-1. This gives fixed points (0,0)(0,0) and (0,−1)(0,-1). Case B: y=0y=0 (and x≠0x \neq 0, to avoid double-counting Case A). The first equation becomes x(3−x)=0x(3-x) = 0; since x≠0x \neq 0 we need 3−x=03 - x = 0, i.e. x=3x = 3. This gives the fixed point (3,0)(3,0). Case C: 3−x−2y=03 - x - 2y = 0 and 2x−1−y=02x - 1 - y = 0 simultaneously (both factors zero, x≠0x\neq0, y≠0y \neq 0). From the second equation, y=2x−1y = 2x-1. Substitute into the first:

3−x−2(2x−1)=0  ⟹  3−x−4x+2=0  ⟹  5−5x=0  ⟹  x=1.3 - x - 2(2x-1) = 0 \;\Longrightarrow\; 3 - x - 4x + 2 = 0 \;\Longrightarrow\; 5 - 5x = 0 \;\Longrightarrow\; x = 1.

Then y=2(1)−1=1y = 2(1)-1 = 1. This gives the fixed point (1,1)(1,1), and indeed x=1≠0x=1\neq0, y=1≠0y=1\neq0 as required for this case. We must also check the mixed cases (x=0x=0 together with 2x−1−y=02x-1-y=0, or y=0y=0 together with 3−x−2y=03-x-2y=0) are not missed — but these are already subsumed: x=0x=0 with 2x−1−y=02x-1-y=0 gives y=−1y=-1, already found in Case A (since x=0x=0 automatically satisfies the first factor); and y=0y=0 with 3−x−2y=03-x-2y=0 gives x=3x=3, already found in Case B. So no fixed points are missed. Collecting all cases, the system has exactly four fixed points:

(0,0),(0,−1),(3,0),(1,1).(0,0), \quad (0,-1), \quad (3,0), \quad (1,1).

Direct verification of (1,1)(1,1): x˙=1(3−1−2)=1(0)=0\dot x = 1(3-1-2) = 1(0)=0 ✓, y˙=1(2−1−1)=1(0)=0\dot y = 1(2-1-1)=1(0)=0 ✓. Both vanish as required. By the theorem, the constant trajectory (x,y)≡(1,1)(x,y)\equiv(1,1) is an exact solution for all t∈Rt \in \mathbb{R}; biologically, this is the coexistence equilibrium of the two populations, while (0,0)(0,0), (0,−1)(0,-1), and (3,0)(3,0) are boundary or extinction-type equilibria (note (0,−1)(0,-1) has a negative population coordinate, so it may be discarded as biologically meaningless even though it is a legitimate mathematical fixed point of the ODE system — a reminder that the ambient domain DD chosen for the model matters).

CounterexampleDropping continuity of f\mathbf{f} breaks the correspondence

The theorem's forward direction (⇒\Rightarrow: f(x∗)=0\mathbf{f}(\mathbf{x}^*)=0 implies the constant function solves the ODE) used continuity of f\mathbf{f} at x∗\mathbf{x}^* to conclude f(x(t))=f(x∗)\mathbf{f}(\mathbf{x}(t)) = \mathbf{f}(\mathbf{x}^*) for every tt. If f\mathbf{f} is allowed to be discontinuous, the correspondence between algebraic zeros of f\mathbf{f} and genuine constant solutions of a meaningful differential equation can become vacuous or misleading. Consider the scalar (badly-behaved) vector field on D=RD=\mathbb{R}:

f(x)={0x=01x≠0.f(x) = \begin{cases} 0 & x = 0 \\ 1 & x \neq 0.\end{cases}

Algebraically f(0)=0f(0) = 0, so x∗=0x^*=0 is a "fixed point" by the bare zero-condition. Our proof's direction (⇒\Rightarrow) still literally goes through for the constant function x(t)≡0x(t)\equiv 0 itself (since x(t)≡0x(t) \equiv 0 trivially forces f(x(t))=f(0)=0=x˙(t)f(x(t))=f(0)=0=\dot x(t) for all tt), so in this toy scalar case nothing breaks yet. The real failure surfaces when we ask for the converse-type consequence normally expected alongside such theorems: that fixed points govern nearby dynamics continuously, i.e. that solutions starting near x∗=0x^*=0 actually behave consistently with ff near 00. Here any solution started at x0≠0x_0\ne 0 must satisfy x˙=1\dot x = 1 (since f≡1f\equiv 1 off the origin), giving x(t)=x0+tx(t) = x_0+t, which reaches x=0x=0 in finite time t=−x0t=-x_0 and then, by the differential equation, would need x˙(0−)\dot x(0^-) to jump discontinuously against f(0)=0f(0)=0 — no differentiable solution can actually satisfy x˙=f(x)\dot x = f(x) through this pieced-together path, and in fact no solution passing through a punctured neighborhood of 00 solves the ODE at all, because differentiability of x(t)x(t) forces x˙(t)=1\dot x(t)=1 arbitrarily close to t=−x0t=-x_0 while the defining equation demands x˙→0\dot x \to 0 as x(t)→0x(t)\to 0, a contradiction. The existence theory simply collapses: the discontinuous ff does not admit continuously differentiable solutions threading the equilibrium at all, so the clean correspondence "zero of ff ⇔\Leftrightarrow constant solution, and nothing else passes through it" degenerates into a statement about an equation that barely has solutions to speak of near that point. This shows why continuity of f\mathbf{f} (used explicitly in direction ⇒\Rightarrow) is not a cosmetic hypothesis: without it, the very notion of "solving the ODE" near the candidate equilibrium can fail to be well-posed, and the clean logical structure of the theorem, which relies on evaluating f\mathbf{f} along a solution curve, no longer connects to genuine trajectories in any neighborhood of the equilibrium.

Scope, edge cases & common traps

SituationWhat to watch for
Non-isolated equilibriaIf f\mathbf{f} vanishes on an entire curve or line (e.g. x˙=y2,y˙=0\dot x = y^2, \dot y = 0 vanishes for all y=0y=0... check carefully), fixed points form a continuum, not a discrete set; "isolated" must be checked, not assumed.
Solving component-wiseNever solve f1=0f_1=0 and f2=0f_2=0 separately and freely mix roots; you must find (x,y)(x,y) satisfying both equations at once, as in the worked examples.
Domain restrictionsA fixed point of the bare algebra may lie outside the physically or mathematically meaningful domain DD (e.g. negative population), as seen with (0,−1)(0,-1) in Example 2.
Complex/extraneous rootsPolynomial elimination (as in Example 1) can introduce extraneous or complex roots; always verify candidates in the original equations.
Confusing f=0\mathbf{f}=\mathbf{0} with x˙\dot{\mathbf{x}} smallA slow trajectory is not an equilibrium; only exact vanishing of the vector field counts.

The mathematician behind it

Origins · 1 minHenri Poincaré and how this was discovered
HP
Henri Poincaré
1854–1912 · Qualitative theory of differential equations, celestial mechanics

Henri Poincaré founded the qualitative and geometric study of differential equations, shifting attention from explicit closed-form solutions to the global structure of phase space: equilibria, periodic orbits, and their stability. His work on the three-body problem showed that locating and classifying fixed points, rather than solving equations in closed form, is often the only tractable route to understanding a nonlinear system's long-term behavior.

History
- 1854 — Henri Poincaré born in Nancy, France.
- 1881–1886 — Poincaré publishes his memoirs on curves defined by differential equations, introducing the classification of equilibrium points (nodes, saddles, foci, centers).
- 1890 — His prize memoir on the three-body problem lays groundwork for what becomes chaos theory, built on phase-space and fixed-point reasoning.
- 1892–1899 — Les méthodes nouvelles de la mécanique céleste develops the qualitative theory of dynamical systems in full generality.
- 1912 — Poincaré dies in Paris; his geometric viewpoint on equilibria and phase portraits becomes the standard language of nonlinear dynamics.
Is every zero of the vector field automatically a stable equilibrium?

No. A fixed point is defined purely by f(x∗)=0\mathbf{f}(\mathbf{x}^*) = \mathbf{0}; this says nothing yet about whether nearby trajectories approach or flee from it. Stability is a separate, finer question addressed once you linearize the system near x∗\mathbf{x}^* using the Jacobian matrix, as developed in Linearization and the Jacobian and formalized in Stability Concepts and Definitions. A fixed point can be stable (a sink), unstable (a source), a saddle, or even display more delicate borderline behavior (a center) depending on the eigenvalues of the linearization. Locating fixed points is always the necessary first step, but it is only the first step.

Can a system with no fixed points still be analyzed qualitatively?

Yes. Some systems, especially those with periodic-only or unbounded flows (for instance x˙=1,y˙=1\dot x = 1, \dot y = 1, which never has f=0\mathbf{f}=\mathbf{0}), have no equilibria at all, and the whole phase space is swept along by a nonzero flow. In such cases the analysis shifts to studying invariant sets like limit cycles, or asymptotic behavior at infinity, rather than equilibria. The absence of fixed points is itself an important qualitative fact, often the first thing one checks and rules out.

How does this generalize from the single-variable phase-line picture?

For a scalar autonomous equation x˙=f(x)\dot x = f(x), fixed points are simply the real roots of ff, plotted as marked points on a one-dimensional phase line, with arrows between them showing the sign of ff. The nn-dimensional case is the direct generalization: instead of a single equation f(x)=0f(x)=0, you solve the simultaneous system f1(x)=⋯=fn(x)=0f_1(\mathbf{x})=\cdots=f_n(\mathbf{x})=0. The conceptual content — "equilibrium equals zero velocity equals constant solution" — carries over unchanged; only the algebra of solving for the zero set becomes genuinely harder in more than one dimension.

Why must uniqueness (Lipschitz continuity) be assumed to conclude that no other trajectory passes through a fixed point?

Without a Lipschitz-type condition, the initial value problem may have multiple solutions through the same point, and one of those extra solutions could pass through a fixed point non-constantly (the classic example being x˙=x2/3\dot x = x^{2/3}, or, adapted appropriately, systems with non-Lipschitz nonlinearities). The Picard–Lindelöf theorem is precisely the tool that rules this out, so its hypothesis (local Lipschitz continuity) must be assumed explicitly, exactly as stated in the theorem, rather than silently used.

Do fixed points have to be real numbers, or can complex solutions of the algebraic system count?

For the differential-equations interpretation to make sense, we always work in the real ambient space D⊆RnD \subseteq \mathbb{R}^n where the dynamical system is actually defined, so only real solutions of f(x)=0\mathbf{f}(\mathbf{x})=\mathbf{0} lying in DD count as fixed points. When elimination produces a polynomial with complex roots (as could happen in principle in examples like the ones above), those complex roots are discarded; they are artifacts of the algebraic method, not points of the real phase space.

What is the difference between "fixed point," "equilibrium point," "critical point," and "steady state"?

In the context of an autonomous ODE system, these four terms are used essentially interchangeably in the literature to mean the same object: a point where the vector field vanishes. "Equilibrium" and "steady state" are more common in physics and engineering contexts emphasizing the constant, unchanging nature of the corresponding solution; "fixed point" is standard in dynamical systems terminology (and also used, with a related but distinct meaning, for discrete maps xn+1=g(xn)\mathbf{x}_{n+1}=g(\mathbf{x}_n)); "critical point" is more common in classical ODE textbooks. All four names refer to the identical mathematical condition f(x∗)=0\mathbf{f}(\mathbf{x}^*)=\mathbf{0} used throughout this article.

How is this used before we know anything about stability?

Locating fixed points is purely algebraic and requires no stability theory at all: it is solving f(x)=0\mathbf{f}(\mathbf{x})=\mathbf{0}, a system of (typically polynomial or otherwise explicit) equations. This makes it the natural, self-contained first step in analyzing any autonomous system, usable immediately after learning what an autonomous system and its phase space are, and entirely independent of the subsequent, more delicate question of classifying the equilibrium's stability via linearization.

Before this

Where this leads