Phase lines, equilibrium stability, bifurcations, nonlinear population models, threshold phenomena, and an introduction to chaos — the tools for understanding systems that cannot always be solved exactly
Not every differential equation yields a tidy closed-form solution, and in many of the most important models in science, engineering, ecology, and economics, the equations are nonlinear. This chapter teaches students to work with equations that cannot always be solved exactly by developing tools for describing what solutions do rather than only what solutions are.
Students will study autonomous equations, phase lines, equilibrium solutions, stability, long-term behavior, bifurcations as an introduction, nonlinear population models, threshold phenomena, and a preview of nonlinear systems and chaos. By the end of this chapter, students should understand that qualitative analysis is not a substitute for exact methods but a complement to them, and that understanding the behavior of a model is often more valuable than a formula that only works under idealized conditions.
A wildlife biologist is tracking a fish population in a lake. The population grows when it is small, but fishing pressure increases as the population grows, and the habitat has limited resources. She writes a differential equation to model this interaction but cannot solve it exactly because of its nonlinear terms. She still needs to answer real questions: Will the population stabilize? Is there a minimum size below which the population will collapse? How does the long-term population depend on current conditions? These questions are not about finding a formula. They are about understanding the behavior of a system. Qualitative analysis is the set of tools that makes this kind of thinking possible.
The methods developed in Chapters 3 through 11 are powerful, but each depends on special structure. Separable equations require the right-hand side to factor as a product of a function of x and a function of y. Linear equations require the unknown function and its derivatives to appear only to the first power, never multiplied together. Characteristic equations depend on constant coefficients. Laplace transforms work beautifully on equations with specific forcing functions.
A large portion of the differential equations that arise in actual science, engineering, ecology, and economics are not linear. As soon as a model includes realistic feedback, saturation, competition, or interaction between quantities, the equation becomes nonlinear. These equations rarely have closed-form solutions. This does not mean the equations are useless. It means that the approach must change.
A differential equation is nonlinear if the unknown function or its derivatives appear in any way that is not linear. This includes products of the unknown function with itself or with its derivatives, exponentials or logarithms of the unknown function, or any other combination that cannot be written as a linear combination of y, y', y'', and so on.
| Equation | Why Nonlinear | Appears In |
|---|---|---|
dy/dt = y(1 − y) | Contains y² when expanded | Logistic population growth |
dy/dt = sin(y) | sin(y) is not linear in y | Nonlinear pendulum |
y'' + sin(y) = 0 | Nonlinear second-order pendulum | Large-angle pendulum |
dy/dt = y² − t | Contains y² | Riccati-type equations |
When an exact solution is unavailable, qualitative analysis asks: what can we determine about the behavior of solutions without finding the solution explicitly? For a first-order autonomous equation dy/dt = f(y), we can locate equilibria, determine where solutions increase or decrease, classify long-term behavior, and analyze how behavior changes with parameters, all without integrating the equation.
One of the most important properties of linear differential equations is the superposition principle: if y₁ and y₂ are both solutions to a homogeneous linear equation, then c₁y₁ + c₂y₂ is also a solution for any constants c₁ and c₂. This principle underlies the entire approach to second-order equations in Chapters 6 through 8 and to systems in Chapter 9.
Nonlinear equations do not satisfy superposition. If y₁ and y₂ are both solutions to a nonlinear equation, their sum y₁ + y₂ is generally not a solution. The solution set cannot be built by adding simpler pieces together.
Solutions to linear equations with bounded coefficients exist for all time. Nonlinear equations can produce solutions that become infinite in finite time. For the equation dy/dt = y² with initial condition y(0) = 1, the solution is y(t) = 1/(1 − t). This solution exists for t < 1 but becomes infinite as t → 1⁻. A solution that starts at y = 1 at t = 0 blows up in exactly one unit of time.
Superposition — holds for linear equations; fails for nonlinear equations.
Finite-time blowup — impossible for linear equations with bounded coefficients; possible for nonlinear equations.
Sensitivity to initial conditions — mild in linear systems; can be extreme in nonlinear systems.
Exact solutions — structured methods often succeed for linear equations; nonlinear equations require qualitative or numerical methods in most cases.
A first-order differential equation is autonomous if the right-hand side depends only on the dependent variable, not on the independent variable:
dy/dt = f(y)
The word autonomous means self-governing. In an autonomous equation, the rate of change depends only on the current state y, not on the current time t. The equation does not change as time passes. It describes a system whose rules are fixed.
For an autonomous equation, the slope of the solution curve at any point depends only on the height y, not on the horizontal position t. The slopes along any horizontal line y = c are all the same regardless of t. This means the entire slope field can be summarized by a single vertical diagram called the phase line.
For dy/dt = f(y), an equilibrium solution is a constant solution y = y* such that f(y*) = 0. Between equilibria, the sign of f(y) determines whether solutions increase or decrease.
y < 0: (−)(+) = − → dy/dt < 0, solution decreasing
0 < y < 3: (+)(+) = + → dy/dt > 0, solution increasing
y > 3: (+)(−) = − → dy/dt < 0, solution decreasing
A phase line is a diagram that summarizes the qualitative behavior of an autonomous first-order differential equation. Instead of drawing the full slope field in two dimensions, the phase line collapses everything onto a single vertical axis representing y. Equilibrium points are marked as solid dots. Upward arrows show where f(y) > 0 and solutions increase. Downward arrows show where f(y) < 0 and solutions decrease.
Step 1 — Find all equilibria by solving f(y) = 0.
Step 2 — Mark the equilibria on a vertical number line.
Step 3 — Choose a test value in each interval between equilibria.
Step 4 — Evaluate the sign of f at each test value. Draw upward arrows where f > 0, downward arrows where f < 0.
Step 5 — Interpret the stability of each equilibrium from the arrow directions.
Equilibria: y = 0 and y = 3
Interval Test value f(y) = y(3−y) Sign Arrow
y < 0 y = −1 (−1)(4) = −4 − Downward
0 < y < 3 y = 1 (1)(2) = 2 + Upward
y > 3 y = 4 (4)(−1) = −4 − Downward
⬤ Gold dot — stable equilibrium (y = 3): arrows from both sides point toward it.
⬤ Red dot — unstable equilibrium (y = 0): arrows from both sides point away from it.
Blue arrow up — f(y) > 0, solutions increasing (between y = 0 and y = 3).
Green arrow down — f(y) < 0, solutions decreasing (above y = 3 and below y = 0).
An equilibrium solution y = y* is called stable if solutions that start near y* remain near y* and approach y* as t → ∞. An equilibrium is called unstable if solutions that start near y* move away from y* as t increases. These definitions are intuitive: a stable equilibrium attracts nearby solutions, while an unstable equilibrium repels them.
| Arrow Pattern | Classification | Behavior |
|---|---|---|
| Both arrows point toward equilibrium | Stable | Nearby solutions converge to y* |
| Both arrows point away from equilibrium | Unstable | Nearby solutions diverge from y* |
| One arrow toward, one arrow away | Semistable | Attracts from one side, repels from the other |
If y* is an equilibrium of dy/dt = f(y), then:
f'(y*) < 0 — the equilibrium is stable.
f'(y*) > 0 — the equilibrium is unstable.
f'(y*) = 0 — the test is inconclusive. Use sign analysis.
Why it works: near y*, write y = y* + u where u is a small perturbation. Then du/dt ≈ f'(y*)·u. If f'(y*) < 0, the perturbation decays and the system returns to y*. If f'(y*) > 0, the perturbation grows.
Equilibria: y² = 4 → y = 2 and y = −2
f'(y) = 2y
At y = 2: f'(2) = 4 > 0 → Unstable
At y = −2: f'(−2) = −4 < 0 → Stable
y < −2: (+)(−) = + (upward) → toward y = −2 ✓
−2 < y < 2: (−)(+) = − (downward) → toward y = −2 ✓
y > 2: (+)(+) = + (upward) → away from y = 2 ✓
A semistable equilibrium occurs when the function f(y) touches zero at y* without changing sign. It has a double zero at y*. Solutions approaching from one side converge toward the equilibrium, while solutions departing from the other side move away. When f'(y*) = 0, the linearization test is inconclusive and sign analysis must be used.
Equilibrium: y = 2 (double zero)
f'(y) = −2(y − 2) → f'(2) = 0 (inconclusive)
Sign analysis:
y < 2: (y−2)² > 0, so f(y) = −(y−2)² < 0. Arrow downward (away from y = 2 going down).
y > 2: (y−2)² > 0, so f(y) < 0. Arrow downward (toward y = 2 from above).
Semistable equilibria are not rare curiosities. They appear in models where a threshold separates attraction from repulsion, such as threshold population models and critical load problems in engineering.
Given an autonomous equation and an initial condition, the phase line allows prediction of long-term behavior without solving the equation. The possible long-term behaviors for a first-order autonomous equation are: (1) the solution approaches a stable equilibrium, (2) the solution diverges to +∞ or −∞, or (3) the solution remains at an equilibrium forever. There is no oscillation in scalar first-order autonomous equations. Oscillation requires at least two interacting quantities.
Interval Test f(y) = y(3−y)(y−1) Sign Arrow
y < 0 −1 (−1)(4)(−2) = −8 − Down
0 < y < 1 0.5 (0.5)(2.5)(−0.5) = −0.625 − Down
1 < y < 3 2 (2)(1)(1) = 2 + Up
y > 3 4 (4)(−1)(3) = −12 − Down
Stability:
y = 0: arrow below points down (away), arrow above points down (toward) → semistable
y = 1: arrow below points down (away), arrow above points up (away) → unstable
y = 3: arrow below points up (toward), arrow above points down (toward) → stable
Many differential equations contain a parameter, a number that can be adjusted to represent different physical conditions. A bifurcation is a qualitative change in the behavior of solutions that occurs at a specific parameter value. At a bifurcation point, the number or type of equilibria changes and the phase line changes its structure.
Consider the family dy/dt = y² + r where r is a parameter.
| Parameter Value | Equilibria | Behavior |
|---|---|---|
r > 0 | None | All solutions increase without bound (f(y) > 0 everywhere) |
r = 0 | y = 0 (semistable) | Bifurcation point: two equilibria collide and disappear |
r < 0 | y = √(−r) (unstable), y = −√(−r) (stable) | Two equilibria, one stable and one unstable |
A bifurcation diagram plots equilibrium values y* on the vertical axis against the parameter r on the horizontal axis. Stable equilibria appear as solid curves; unstable equilibria as dashed curves. For the saddle-node example, two curves meet at the bifurcation point (r = 0, y* = 0), forming a parabola opening to the left. For r > 0, no equilibria exist. Reading the diagram vertically at any fixed r gives all equilibria at that parameter value.
f(y) = ry − y² = y(r − y)
Equilibria: y = 0 and y = r
f'(y) = r − 2y
At y = 0: f'(0) = r
r > 0: unstable r < 0: stable
At y = r: f'(r) = −r
r > 0: stable r < 0: unstable
Bifurcations appear throughout science and engineering. A population model can undergo a transcritical bifurcation at a threshold harvesting rate. A physical system can undergo a saddle-node bifurcation as a parameter is tuned. Recognizing bifurcation structure is a critical modeling skill in ecology, economics, physics, and control systems.
Chapter 5 introduced exponential growth dP/dt = kP as a first approximation to population dynamics. Exponential growth predicts unlimited increase, which is unrealistic. The logistic equation incorporates competition and resource limits:
dP/dt = rP(1 − P/K)
where r > 0 is the intrinsic growth rate and K > 0 is the carrying capacity. Equilibria: P = 0 (unstable) and P = K (stable). Populations starting at any positive initial value approach K.
Adding a constant removal rate h yields the harvesting model:
dP/dt = rP(1 − P/K) − h
For small h, two positive equilibria exist. For large h, no positive equilibria exist and the population collapses. The critical value of h at which equilibria appear or disappear is a bifurcation point.
Setting P(1 − P/100) − h = 0 and rearranging:
P² − 100P + 100h = 0
P = (100 ± √(10000 − 400h)) / 2 = 50 ± √(2500 − 100h)
For h = 20:
P = 50 ± √500 ≈ 50 ± 22.4
P₁ ≈ 27.6 (unstable, f'(P₁) > 0)
P₂ ≈ 72.4 (stable, f'(P₂) < 0)
For h = 30:
Discriminant = 2500 − 3000 = −500 < 0. No real equilibria.
dP/dt < 0 for all P > 0 → population collapses.
A threshold in a differential equation model is a critical value of the state variable below which one type of behavior occurs and above which another type occurs. Mathematically, a threshold often corresponds to an unstable or semistable equilibrium.
In some populations, small populations decline due to difficulty finding mates, defending against predators, or maintaining social structure. This is the Allee effect. Below a critical size A, the population declines even without external pressure. A simple model is:
dP/dt = rP(P/A − 1)(1 − P/K) where 0 < A < K
| Equilibrium | Stability | Biological Meaning |
|---|---|---|
P = 0 | Stable | Extinction attractor |
P = A | Unstable | Minimum viable population (threshold) |
P = K | Stable | Carrying capacity |
P(0) = 10 < 20 (below threshold A): dP/dt < 0 → extinction (P → 0)
P(0) = 50 (between A and K): dP/dt > 0 → P → 100 (carrying capacity)
P(0) = 120 > 100 (above K): dP/dt < 0 → P → 100 (carrying capacity)
When two or more quantities interact, a system of differential equations is needed. Chapter 9 introduced linear systems. Nonlinear systems are more complex but also more realistic. The key tool for analyzing them qualitatively is the phase plane.
For a two-equation system dx/dt = f(x, y), dy/dt = g(x, y), the phase plane is the xy-plane. A solution (x(t), y(t)) traces a trajectory. Nullclines are curves where one component is zero: the x-nullcline is where f(x,y) = 0 (trajectory moves vertically), and the y-nullcline is where g(x,y) = 0 (trajectory moves horizontally). Equilibrium points occur at intersections of x- and y-nullclines.
For a nonlinear system, stability at each equilibrium is determined locally by the Jacobian matrix:
J = [[∂f/∂x, ∂f/∂y], [∂g/∂x, ∂g/∂y]] evaluated at the equilibrium point.
The eigenvalues of J govern local stability in the same way as for linear systems (Chapter 9):
Both eigenvalues Re < 0 — stable node or spiral.
Either eigenvalue Re > 0 — unstable (saddle or unstable node/spiral).
Purely imaginary eigenvalues — inconclusive for the nonlinear system (may be center or spiral).
x-nullcline (dx/dt = 0): x = 0 or x + y = 2
y-nullcline (dy/dt = 0): y = 0 or x + 2y = 3
Intersect x=0 with y=0: → (0, 0)
Intersect x=0 with x+2y=3: → (0, 3/2)
Intersect x+y=2 with y=0: → (2, 0)
Intersect x+y=2 with x+2y=3: subtract to get y=1, x=1 → (1, 1)
J = [[ 2−2x−y, −x ],
[ −y, 3−x−4y ]]
At (0,0): eigenvalues λ₁=2, λ₂=3 → Unstable node
At (0,3/2): eigenvalues λ₁=1/2, λ₂=−3 → Saddle (unstable)
At (2,0): eigenvalues λ₁=−2, λ₂=1 → Saddle (unstable)
At (1,1): char. eq. λ²+3λ+1=0, λ = (−3±√5)/2 ≈ −0.38, −2.62 → Stable node
Some nonlinear systems produce trajectories that converge to a closed loop in the phase plane called a limit cycle. Limit cycles represent sustained oscillations that neither grow nor decay. They are inherently nonlinear phenomena and do not occur in linear systems. Limit cycles appear in models of biological rhythms (heartbeat, circadian rhythms), electrical oscillators, and chemical reactions.
In mathematics, chaos has a precise meaning. A dynamical system is chaotic if it exhibits all of the following properties:
| Property | Description |
|---|---|
| Sensitivity to initial conditions | Two solutions starting arbitrarily close together diverge exponentially fast. Small differences in initial conditions lead to completely different long-term behavior. |
| Topological mixing | Solutions that start in different regions of the state space eventually wander near each other. |
| Dense periodic orbits | The phase space contains infinitely many periodic solutions that are densely distributed. |
The first property is the most familiar. It is sometimes called the butterfly effect: the flap of a butterfly's wings in one location might eventually influence a storm on the other side of the world.
Chaotic systems are deterministic. Given an exact initial condition, the future is completely determined by the differential equation. Chaos is not random. But because no initial condition can be measured with perfect precision, and because nearby solutions diverge rapidly, long-term prediction in a chaotic system is practically impossible even though the underlying equation is exact.
Determinism does not guarantee predictability. This is a profound insight.
The most famous chaotic system in differential equations is the Lorenz system, derived by meteorologist Edward Lorenz in 1963 as a simplified model of atmospheric convection:
dx/dt = σ(y − x)
dy/dt = x(ρ − z) − y
dz/dt = xy − βz
For the classical parameter values σ = 10, ρ = 28, β = 8/3, the system produces chaotic behavior. The long-term solution traces a butterfly-shaped object in three-dimensional space called the Lorenz attractor. Solutions never settle to an equilibrium and never repeat, but remain within a bounded region and continuously wind around two lobes.
Draw two oval loops, one on the left and one on the right, slightly overlapping at the center. Draw spiraling trajectories that orbit around the left loop for several turns, then cross to the right loop, orbit there for several turns, then cross back. The number of turns before crossing is unpredictable. Label this the Lorenz attractor. Label the axes x, y, z.
| Field | Where Chaos Appears |
|---|---|
| Weather and climate | Atmospheric models are chaotic; weather forecasts become unreliable beyond about two weeks |
| Population dynamics | Discrete and continuous models can exhibit chaos when growth rates are high |
| Fluid dynamics | Turbulence in fluids is related to chaotic dynamics |
| Electronics | Certain nonlinear circuits exhibit chaotic oscillations |
| Biology | Neural dynamics, cardiac rhythms, and ecological systems can exhibit chaos |
A full treatment of chaotic dynamics requires advanced mathematics including dynamical systems theory, topology, and measure theory. This section is a preview. Students continuing into dynamical systems courses or advanced applied mathematics will encounter chaos more deeply. The key takeaway: deterministic differential equations can produce practically unpredictable behavior, and this unpredictability is a structural feature of nonlinear systems, not a defect in the equations.
Phase lines must be constructed from a careful sign analysis of f(y), not guessing. Always choose a specific test value in each interval and evaluate f explicitly. Intuitive sketching often misses the correct direction, especially near semistable equilibria.
An equilibrium is a value y* where f(y*) = 0. An initial condition is a given starting value y(0) = y₀. Always solve f(y) = 0 first to find equilibria, then locate the initial condition on the phase line separately.
If f'(y*) = 0, the derivative test gives no information. The equilibrium may be stable, unstable, or semistable depending on higher-order behavior. Use sign analysis: determine arrow directions on each side of y* to classify the equilibrium.
First-order scalar autonomous equations cannot produce oscillating solutions. Arrows in each interval point in one direction. Oscillation requires at least two interacting variables or a second-order equation. Phase line analysis predicts only monotone behavior toward equilibria or divergence.
For population models, only P ≥ 0 is physically meaningful. Always identify the domain of biological validity and restrict phase line interpretation to that domain. An equilibrium at a negative value has no biological meaning in a population model.
Read the bifurcation diagram vertically, not horizontally. Fix a parameter value on the horizontal axis and look vertically to find all equilibria at that parameter value. Solid curves are stable; dashed curves are unstable. Following a single curve horizontally conflates different equilibrium branches.
Jacobian eigenvalue analysis provides local information only. It describes behavior very close to the equilibrium. A system may be locally stable at an equilibrium but still have solutions that escape to infinity from far away. Treat eigenvalue classification as a local statement.
Chaotic systems are deterministic. Given an exact initial condition, the differential equation determines the future completely. Unpredictability arises from exponential amplification of tiny differences in initial conditions, not from any randomness in the equations. Chaos is a property of deterministic dynamics.
Chapter 12 has developed the ability to analyze differential equations when exact solutions are unavailable. Phase lines, stability classification, bifurcation analysis, threshold modeling, nonlinear systems, and the preview of chaos all belong to the subject of dynamical systems, the study of how systems change over time.
All of the equations in Chapters 1 through 12 are ordinary differential equations (ODEs). The unknown function depends on a single independent variable. Chapter 13 introduces partial differential equations (PDEs), where the unknown function depends on two or more independent variables. The heat equation describes temperature depending on both time t and position x. The wave equation and Laplace's equation similarly involve multiple variables.
The shift from ODE to PDE represents a qualitative expansion in the kind of system that can be described. Space and time together, rather than time alone, open up the full range of physical phenomena: heat flow, wave propagation, diffusion, electrostatics, fluid flow, and quantum mechanics.
The distinction between initial conditions (values at a starting time) and boundary conditions (values at the edges of a spatial domain) becomes especially important in PDEs. The separation of variables technique previewed in Chapter 13 shares a family resemblance with the separable equations of Chapter 3. The qualitative thinking developed throughout this chapter — asking about long-term behavior, stability, equilibrium, and the structure of solutions — applies just as naturally to PDEs.
For each equation: find all equilibria, determine the sign of f(y) in each interval, construct the phase line, classify each equilibrium as stable, unstable, or semistable, and describe the long-term behavior from the given initial conditions.
Each problem describes a student's reasoning. Identify the error and write the correct analysis.
Identify the error in this reasoning. What would the correct conclusion be if both arrows point upward toward y = 2?
Explain why this conclusion is not justified. What does f'(y*) = 0 actually tell us, and what should the student do instead?
Identify what is missing from this analysis. What initial conditions lead to a different conclusion?
Explain what is wrong with reading the bifurcation diagram this way. How should the bifurcation diagram be read correctly?
Write a careful correction to this statement. What is the precise difference between chaos and randomness?
Consider dy/dt = y³ − 4y = y(y − 2)(y + 2).
For dy/dt = sin(y)cos(y) = ½sin(2y) on the domain −π ≤ y ≤ π:
Consider the family dy/dt = y³ − ry where r is a real parameter.
A population of deer follows the Allee effect model: dP/dt = 0.1P(P/30 − 1)(1 − P/300).
Consider the predator-prey system: dx/dt = x(1 − x) − xy, dy/dt = −y + xy, where x is prey and y is predator.
All of the differential equations studied in Chapters 1 through 12 describe how one quantity changes with respect to a single independent variable. Chapter 13 enters a fundamentally different setting. In many physical, biological, and engineering systems, the unknown quantity depends on both time and space simultaneously. Temperature in a rod depends on position along the rod and on the time elapsed. Concentration of a chemical depends on where in a solution you look and when you measure it. These systems are described by partial differential equations, where derivatives are taken with respect to each independent variable. The heat equation, the wave equation, and Laplace's equation are the three classical PDEs introduced in Chapter 13. The qualitative habits developed in this chapter, including stability thinking, equilibrium analysis, and boundary condition awareness, remain essential in the PDE setting.