EDUKATE · THE GOLD STANDARD
Find your next learning step
Start with the vocabulary, follow the worked examples, or go straight to checking a simulation. The guide separates what a mathematical rule establishes from what observations support about the world.
Full chapter index · Diagnose the difficulty · Specialist reading library · Mathematical Modelling guide
Nonlinear dynamics studies how systems change when their governing relationships are not linear. It asks whether a state settles, oscillates, changes regime or becomes difficult to predict in detail. The subject connects equations with geometry, computation and evidence.
Nonlinear does not mean random, and it does not automatically mean chaotic. A nonlinear system can settle calmly at an equilibrium. A deterministic rule can also produce complicated behaviour. The important task is to identify the rule, state, parameters and conditions before describing the outcome.
This M–N root guide connects the eduKate branches on feedback, stability, oscillation, bifurcation and modelling. Its small mathematical examples are chosen so that a reader can check the reasoning rather than accept a dramatic graph on trust.
The wider aim is learning that transfers into life. The worked systems are educational models, not measurements, forecasts or engineering approvals. Applying them to weather, ecosystems, machines or people requires the relevant subject knowledge and evidence.
Full chapter index
Understand change · Chapters 1–4
Read stability and geometry · Chapters 5–9
Explore repeated rules · Chapters 10–14
Connect the evidence · Chapters 15–18
Check and transfer · Chapters 19–22
A linear relationship preserves the relevant scaling and addition properties. In the simple function y = 3x, doubling x doubles y, and the output for two added inputs is the sum of their separate outputs. The function y = x² does not satisfy those properties.
Nonlinearity in a dynamical model can arise through powers, products, thresholds or other relationships. Its significance comes from how the rule changes motion or updating, not merely from how curved a graph looks. The mathematical definition should come before an everyday metaphor.
Compare the rates x′ = −x and x′ = −x³. Both direct positive x toward smaller values, but their magnitudes depend differently on the state. At x = 2, the rates are −2 and −8; at x = 0.5, they are −0.5 and −0.125. Nonlinearity changes how the response scales.
Begin with What Is Nonlinearity? and Algebra. The first useful distinction is between a property of a rule and a claim about a real system. A model can contain a nonlinear term without proving that it adequately represents the object being studied.
The state is the set of quantities used to describe the system at a given time or step. A parameter specifies part of the governing rule. An observation is what has actually been measured. These are related concepts, not interchangeable names.
In a model of a moving object, position alone may not determine the next motion; velocity can also be needed. In a one-variable recurrence, the current value may be sufficient by construction. The appropriate state description depends on the model.
An initial condition specifies a starting state. Changing it asks what happens from another starting point under the same rule. Changing a parameter changes the rule itself. Keeping that distinction visible is essential when comparing trajectories or drawing a bifurcation diagram.
Use What Is a Control Parameter? and Mathematical Modelling. Record quantities, units, allowed values and the meaning of the time variable. All variables and time in the simple cubic example below are nondimensional, so its coefficients are not being offered as measured physical constants.
A continuous-time model describes a state evolving with time, often using differential equations. A discrete-time map specifies an update from one step to the next. Similar-looking equations in the two settings can have very different behaviour.
The equation x′ = f(x) specifies a rate. The equation xₙ₊₁ = f(xₙ) specifies a next value. A prime on x is not a decorative alternative to a step index. Mixing the two interpretations can produce false conclusions about stability or oscillation.
A numerical method can approximate a continuous flow using discrete updates. In that case, the step is part of the calculation method. It is not necessarily a natural cycle of the represented system, and changing it should not silently change the underlying model.
For the wider mathematical setting, read Dynamical Systems and Calculus. Before running a simulation, say whether you are iterating a map as the model itself or approximating a continuous equation. That single sentence prevents several common errors.
There is a useful structural difference. A smooth autonomous scalar flow on a line cannot have a nonconstant periodic orbit: between equilibria, the sign of its rate fixes the direction of motion. A discrete map can jump between values. The logistic map’s period-two behaviour therefore does not contradict the simpler geometry of a one-variable continuous flow.
CHAPTER 4 OF 22 · Understand change
4. Equilibrium: a possible state, not a guarantee
An equilibrium of x′ = f(x) satisfies f(x*) = 0. If a solution starts exactly at that equilibrium and the model has the usual uniqueness properties, it remains there. Finding equilibria is a first step, not a complete account of motion from other starting points.
For a discrete map xₙ₊₁ = F(xₙ), a fixed point satisfies F(x*) = x*. Setting F(x*) = 0 instead generally asks a different question. The right calculation follows from what the left-hand side represents.
Stability concerns responses to perturbations. Asymptotic stability adds eventual approach to the equilibrium. A state can satisfy the equilibrium equation yet repel nearby starting points. “An equilibrium exists” does not mean “the system will settle there”.
MIT’s First Order Autonomous Differential Equations provides a foundational route into qualitative analysis. The practical question is not only where change stops, but which way nearby states move.
CHAPTER 5 OF 22 · Read stability and geometry
5. Worked flow: two attracting states and one repelling state
Consider the nondimensional teaching model x′ = 2x − x³ = x(2 − x²). Equilibria occur at x = 0 and x = ±√2. The variable x is an abstract signed state, not a population count, so negative values are allowed in this example.
At x = 1, the rate is 1, so x increases. At x = 2, the rate is −4, so x decreases. On the positive side, the flow therefore points toward √2. At x = −1, the rate is −1; at x = −2, it is 4. On the negative side, the flow points toward −√2.
Near zero, positive states move toward larger positive values and negative states toward more negative values. Zero is therefore repelling, while the two nonzero equilibria attract nearby states. Starting exactly at zero remains a separate case.
The example shows why one equation can have different long-run outcomes depending on its initial state. No random choice is needed in this model. It also shows why an average of two possible final states need not itself be an attracting final state: the average of ±√2 is zero, which is repelling here.
CHAPTER 6 OF 22 · Read stability and geometry
6. Local stability: use the right derivative test
For the continuous scalar equation x′ = f(x), a small displacement η from an equilibrium approximately follows η′ = f′(x*)η when linearisation applies. A negative derivative gives local exponential decay of small displacements; a positive derivative gives local growth.
In the cubic example, f′(x) = 2 − 3x². At zero this is 2, agreeing with repulsion. At either ±√2 it is −4, agreeing with local attraction. The derivative calculation supports the direction-of-flow reasoning rather than replacing it with an unexplained label.
For a smooth discrete map, the corresponding first-order update is ηₙ₊₁ ≈ F′(x*)ηₙ. The usual strict local attraction condition is |F′(x*)| < 1, not simply a negative derivative. A multiplier of −2 alternates sign and grows in magnitude.
When the continuous derivative is zero or the map multiplier has magnitude one, these simple first-order tests can be inconclusive. Higher-order behaviour matters. “The test does not decide” is a valid mathematical outcome; it is not permission to guess.
CHAPTER 7 OF 22 · Read stability and geometry
7. Phase portraits and basins of attraction
A phase portrait represents possible trajectories in state space. It shows how states move, rather than only plotting one measured quantity against time. When the state has several components, the axes must identify which components are being shown.
A basin of attraction is the set of starting states whose trajectories approach a particular attractor under the stated dynamics. In the cubic example, positive initial states approach √2 and negative ones approach −√2. The exactly zero state remains at zero.
A projection can hide information. Two trajectories may appear to cross in a plot that omits a state coordinate without violating uniqueness in the full state space. Check what has been projected before treating a picture as a contradiction.
Continue into What Is a Vector Field?, What Is an Attractor? and How Mathematical Stability Works. These routes help distinguish a single time trace, a family of possible motions and the geometry of their destinations.
CHAPTER 8 OF 22 · Read stability and geometry
8. Bifurcations: change the rule, then compare regimes
A bifurcation is a qualitative change in a system’s dynamical structure as a parameter changes. It is not just any large change in an observed number. A different initial condition at a fixed parameter is also not, by itself, a bifurcation.
Consider the family x′ = μx − x³. For μ < 0, zero is the only real equilibrium and attracts nearby states. For μ > 0, zero repels and the equilibria ±√μ appear and attract. This is the standard supercritical pitchfork example.
At μ = 0, the equation is x′ = −x³. The vanishing linear derivative does not mean there is no restoring behaviour: positive states decrease and negative states increase. It illustrates why a zero derivative requires more analysis.
MIT’s Flows and Bifurcations in One Dimension develops the normal form and its symmetry. Continue into What Is a Bifurcation? for the conceptual bridge. A symmetric teaching equation does not establish that every real-world transition has pitchfork structure.
Multistability means that more than one attracting outcome is available under the same parameter setting. Hysteresis is a different claim about dependence on the path through changing conditions. The existence of two attracting states does not by itself establish a hysteresis loop. To investigate one, specify how a parameter is changed, how slowly it is changed and what states are recorded on each route.
CHAPTER 9 OF 22 · Read stability and geometry
9. Oscillations, limit cycles and forcing
An oscillation is repeated variation; its mechanism still needs explanation. A system may oscillate because of its internal dynamics, because it is periodically driven, or through an interaction between the two. Similar time traces do not establish identical mechanisms.
A limit cycle is an isolated periodic orbit of an autonomous continuous-time system. Not every closed orbit is a limit cycle. For example, a family of neighbouring periodic orbits around a neutral centre does not consist of isolated attracting cycles.
Adding an external periodic input changes what must be specified. Include its amplitude, frequency and phase where relevant. To represent a periodically forced system as autonomous, the forcing phase can be included as an additional state.
The linked routes are Oscillation, Phase, Resonance, Coupling and Synchronisation. These words describe different relationships. Synchronisation concerns coordinated timing; resonance concerns response to forcing. Neither word is a synonym for chaos.
For a concrete distinction, imagine recording a pendulum’s angle. The same angle can occur while moving in opposite directions, so angle alone may not be a sufficient state. A phase portrait using angle and angular velocity distinguishes those cases. Sampling only once per forcing cycle gives another kind of record, called a stroboscopic view; its repeated points must be interpreted with that sampling convention in mind.
CHAPTER 10 OF 22 · Explore repeated rules
10. The logistic map: a small nonlinear update rule
Consider xₙ₊₁ = rxₙ(1 − xₙ), with 0 ≤ x₀ ≤ 1 and 0 ≤ r ≤ 4. Within these assumptions the map sends the interval [0, 1] into itself. The variable is nondimensional; this is a teaching model, not a calibrated population forecast.
The fixed-point equation is x* = rx*(1 − x*). It has x* = 0 and, for r ≠ 0, x* = 1 − 1/r. The second fixed point lies in the positive part of the interval when r > 1.
The derivative is F′(x) = r(1 − 2x). At the positive fixed point it becomes 2 − r. The strict local attraction condition |2 − r| < 1 gives 1 < r < 3. This does not classify the boundary cases using the strict test alone.
At r = 2, the positive fixed point is 0.5 and its multiplier is zero. At r = 3.2, the fixed point is 0.6875 and its multiplier is −1.2, so it is unstable. The equation still has the fixed point even when nearby trajectories do not approach it.
CHAPTER 11 OF 22 · Explore repeated rules
11. Worked iteration: repeating values need not be equilibrium
Set r = 3.2 and x₀ = 0.2. The first update is 3.2 × 0.2 × 0.8 = 0.512. The second is approximately 0.799539. Continuing the same rule gives the values in the table.
| Step n | xₙ, rounded for display |
|---|---|
| 0 | 0.200000 |
| 1 | 0.512000 |
| 2 | 0.799539 |
| 3 | 0.512884 |
| 4 | 0.799469 |
| 5 | 0.513019 |
| 6 | 0.799458 |
The trajectory approaches alternating values near 0.513045 and 0.799455, rather than the unstable fixed point 0.6875. A period-two orbit satisfies F(F(x)) = x while not remaining fixed after a single update.
Observing a few alternating values suggests a pattern; analysis is needed to establish the cycle and its stability. The table is a worked calculation, not proof about every possible initial condition. In particular, starting exactly at a fixed point remains a special case.
This is one reason not to confuse “the rule is deterministic” with “the state eventually becomes constant”. A completely specified deterministic update can approach a repeating sequence instead.
The period-two values can also be checked algebraically. For the logistic map they are [r + 1 ± √((r − 3)(r + 1))] ÷ (2r), when the two distinct real points exist. At r = 3.2 they give approximately 0.5130445 and 0.7994555. The two-step multiplier is F′(p)F′(q) = −r² + 2r + 4 = 0.16. Its magnitude is below one, supporting local attraction to this cycle under the smooth-map test. This adds analytical evidence to the finite table.
CHAPTER 12 OF 22 · Explore repeated rules
12. Chaos is not just a complicated-looking graph
Chaos concerns deterministic dynamics with irregular long-term behaviour and sensitive dependence on initial conditions, with the precise definition depending on the mathematical setting. Sensitivity by itself is not sufficient: simple unbounded exponential growth can also separate nearby starting values.
Do not diagnose chaos from visual roughness alone. Measurement noise, external forcing, changing parameters or numerical errors can all make a trace look irregular. A real-data claim needs suitable methods and consideration of these alternatives.
MIT’s Chaos and Non-Linear Dynamics chapter develops the distinction between deterministic evolution and predictability. Lorenz’s 1963 paper Deterministic Nonperiodic Flow is a foundational primary research route.
Chaos does not mean that every question becomes unanswerable. A precise long-horizon trajectory, a short-horizon estimate and a long-run distribution are different targets. The evidence required and the predictability of each target can differ.
CHAPTER 13 OF 22 · Explore repeated rules
13. Worked sensitivity: nearby starts can separate
Iterate the logistic map at r = 4 from two starts, 0.2 and 0.200001. They differ initially by one millionth. No random input is added. The table shows selected outputs from an ordinary finite-precision calculation.
| Step | Start 0.2 | Start 0.200001 |
|---|---|---|
| 0 | 0.200000 | 0.200001 |
| 5 | 0.585421 | 0.585381 |
| 10 | 0.147837 | 0.148746 |
| 20 | 0.820014 | 0.031369 |
By step 20 these particular computed trajectories are substantially separated. This demonstrates the finite calculation shown; the table alone is not a mathematical proof of chaos. Longer numerical trajectories are also affected by floating-point precision.
The separation need not increase at every step. Bounded dynamics can stretch and fold differences, so a single-step decrease in separation does not settle the long-term question. Conversely, a short interval of increasing separation is not enough to establish a chaotic regime.
Keep exceptional initial conditions in view. At r = 4, x₀ = 0.25 maps exactly to 0.75, which is a fixed point. That does not make the fixed point stable. It shows why “every trajectory behaves irregularly” would be too strong.
A conditional prediction-horizon calculation
If a small separation is approximated over a relevant interval by δ(t) ≈ δ₀e^(λt), a positive λ describes exponential separation on that scale. Lyapunov exponents formalise growth rates of infinitesimal perturbations along trajectories; estimating them reliably is more demanding than fitting any short rising segment.
For a deliberately hypothetical constant rate λ = 0.2 per time unit, initial uncertainty δ₀ = 0.001 and tolerance 0.1, the approximation reaches the tolerance after ln(100) ÷ 0.2, about 23.0 time units. Halving the initial uncertainty adds ln(2) ÷ 0.2, about 3.47 units, rather than doubling the horizon. The arithmetic is conditional on the exponential approximation remaining appropriate; it is not a universal forecast limit.
CHAPTER 14 OF 22 · Explore repeated rules
14. Numerical instability can imitate a dynamical discovery
Consider the continuous equation x′ = −x with x(0) = 1. Its exact solution e⁻ᵗ decays smoothly toward zero. Although this equation is linear, it provides a useful warning when studying nonlinear simulations.
Forward Euler with step h gives xₙ₊₁ = xₙ − hxₙ = (1 − h)xₙ. With h = 0.5 the factor is 0.5, and the computed sequence decays. With h = 2.5 the factor is −1.5, so the sequence alternates and grows in magnitude.
The growing oscillation at h = 2.5 is a property of this numerical approximation, not of the original differential equation. A striking graph could therefore be an implementation or discretisation problem rather than a new physical effect.
Use Numerical Analysis to distinguish model behaviour from approximation behaviour. Compare step sizes, methods and appropriate reference cases. A simulation’s visual smoothness or complexity is not a substitute for checking the computation.
CHAPTER 15 OF 22 · Connect the evidence
15. Parameters must be estimated and the model tested
Knowing how a mathematical family behaves does not establish which family describes an observed system. A real application needs defensible state variables, parameter information, observations and a validation question.
Distinguish changing the starting estimate from changing the mechanism. If many parameter combinations reproduce the observed output, the parameters may not be identifiable from those data. A fitted curve does not automatically reveal an internal cause.
Separate calibration from evaluation. Data used repeatedly to tune a model cannot provide the same kind of independent challenge as a relevant new test. The test should match the intended forecast horizon, regime and output.
Follow Mathematical Modelling, Statistics and Digital Twins, State Estimation and Parameter Calibration. Updating a state estimate and validating a governing rule are related but distinct activities.
A useful fictional audit starts with an irregular experimental trace and a nonlinear model that resembles it. The next question is not whether the two pictures feel alike. Compare defined outputs under recorded conditions, examine measurement error, and test whether a simpler stochastic or externally driven explanation also accounts for the observations. Good modelling should expose competing explanations and the evidence needed to distinguish them.
CHAPTER 16 OF 22 · Connect the evidence
16. Applications need the subject as well as the equation
In Mechanics and Dynamics, forces, motion and constraints determine which equations are appropriate. In Meteorology, physical processes and observations constrain forecasts. A generic nonlinear equation cannot replace either foundation.
In Ecology, growth and interaction models require biological interpretation. In Networks, connections can influence collective behaviour. Similar mathematical structures can be useful across subjects, but their parameters and assumptions do not transfer automatically.
A responsible analogy identifies both the shared structure and the point where the analogy stops. A pitchfork sketch can explain a possible form of transition without proving that a particular organisation, market or person follows that normal form.
The useful transfer is a method of inquiry: define the state, identify feedback, check possible regimes, test sensitivity and compare with evidence. It is not a licence to label every sudden change a bifurcation or every difficult forecast an example of chaos.
CHAPTER 17 OF 22 · Connect the evidence
17. Local approximations are useful within their limits
Linearisation studies a nonlinear system near a chosen state. It can reveal local behaviour and simplify calculation. It is not an assertion that the entire system is linear or that a distant trajectory will behave the same way.
In the cubic teaching example, the linear approximation near zero predicts repulsion. It does not include the eventual attraction to ±√2 because that requires the nonlinear structure away from zero. A locally correct approximation can be incomplete globally.
State the neighbourhood, regime and purpose of an approximation. If the question concerns movement between basins or a parameter crossing a critical value, a single local model may not cover the whole path.
Continue into Nonlinear Analysis and Calculus. Ask what an approximation preserves, what it omits and which observation would signal that the omitted behaviour matters. “Useful nearby” and “valid everywhere” are very different claims.
CHAPTER 18 OF 22 · Connect the evidence
18. Your connected nonlinear-dynamics library
For foundations, use Mathematics, Algebra, Calculus and Mathematical Modelling. The mathematical routes are Dynamical Systems, Nonlinear Analysis and Mathematical Stability.
For the core concepts, follow Nonlinearity, Feedback, Vector Fields, Attractors, Control Parameters and Bifurcations. These pages help distinguish the ingredients of an explanation rather than using the terms as interchangeable metaphors.
For rhythmic behaviour, continue into Oscillation, Phase, Resonance, Coupling and Synchronisation. For a specialist wave example, read How a Wave Can Travel Without Spreading Out.
For computation and uncertainty, use Numerical Analysis, Probability, Statistics and Counterfactual Stability Reasoning. Keep uncertainty about observations separate from sensitivity within an assumed model.
For applications, return to Physics, Mechanics, Meteorology, Ecology and Networks. This is a curated map of relevant branches, not a claim that every application has been listed.
CHAPTER 19 OF 22 · Check and transfer
19. Diagnose the misunderstanding before adding complexity
If a learner says nonlinear means chaotic, use the cubic flow: trajectories can approach an equilibrium without irregular motion. If they say deterministic means predictable forever, use the nearby-start calculation while explaining its limited evidential role.
If a learner finds the wrong fixed point, ask whether the equation gives a rate or a next value. For a flow, set the rate to zero. For a map, equate the next value with the current value. Require an interpretation of the equation before algebra.
If a learner labels a fixed point stable because its derivative is negative, ask whether they are analysing a flow or a map. The map multiplier −2 exposes why the continuous-time sign rule cannot simply be transferred.
If a simulation suddenly oscillates or grows, first check the method, step size, update order and a simple reference case. The aim is not to explain away interesting results; it is to establish which system—the intended model or the numerical method—produced them.
Is every nonlinear system chaotic?
No. Nonlinear systems can approach equilibria, follow periodic motion or show other behaviour. Chaos is a more specific dynamical claim.
Is chaos the same as randomness?
No. A deterministic rule can produce chaotic dynamics without adding random input. Observational noise and stochastic mechanisms are separate possibilities.
Does an equilibrium have to attract nearby states?
No. An equilibrium can be unstable. Existence and stability are different questions.
Can one time-series plot prove chaos?
Not by appearance alone. Alternative explanations, measurement conditions and analysis methods matter. A short illustrative numerical table is also not a proof of a general mathematical property.
Why can the continuous logistic equation and logistic map differ?
One specifies a continuous rate and the other a discrete update. They are different dynamical rules, not simply two notations for the same process.
What is the safest starting point for a beginner?
Define the state and rule, calculate equilibria, check directions or local stability, and reproduce a small example by hand before interpreting an elaborate simulation.
CHAPTER 21 OF 22 · Check and transfer
21. A four-week route from vocabulary to analysis
Week one: build a quantity dictionary for each teaching model. Identify its state, parameter, time convention and initial condition. Practise saying whether an equation gives a rate or an update without solving it.
Week two: analyse the cubic flow. Find its equilibria, test the sign of the rate between them, and compare with the derivative calculation. Change the initial sign and explain why the destination changes.
Week three: reproduce the logistic-map table using unrounded stored values. Compare r = 2 with r = 3.2. State which observations are numerical illustrations and which conclusions follow from algebra or an established stability test.
Week four: audit the Euler example and explain why changing a numerical step can create a misleading graph. Then solve a changed problem and teach the distinction to someone else. The learning standard follows the Secondary 1 Mathematics learning guide: secure the meaning, practise the method, check the reasoning and transfer it to a new case.
CHAPTER 22 OF 22 · Check and transfer
22. Primary learning routes and the final standard
All numerical examples here are educational calculations under stated rules. They are not measurements of a physical system. Rounded displays are for reading; repeated updates use the stored values unless stated otherwise.
MIT’s First Order Autonomous Differential Equations is the foundational route. Its Flows and Bifurcations in One Dimension develops phase-line analysis and elementary bifurcations.
MIT’s Chaos and Non-Linear Dynamics provides a deeper mechanics-oriented treatment. The Nonlinear Dynamics and Chaos syllabus maps a more advanced course; it assumes differential-equation foundations.
Lorenz’s Deterministic Nonperiodic Flow is an original research reference, not a ready-made forecast for a particular location. Keep historical significance, mathematical analysis and current empirical application distinct.
The final standard is a connected explanation: this state evolves by this rule; these equilibria or regimes follow; these checks support the calculation; these observations support the application; these limits remain. Nonlinear dynamics is useful when it sharpens a question instead of replacing it with a dramatic label.
Contents · Previous chapter · Learning that transfers into life