Differential Equations › Section E · Systems of differential equations › Topic E.13
Applications of nonlinear dynamics
Equilibria, linearisation, limit cycles and conserved quantities are not just theory. They explain why predator and prey numbers rise and fall in turn, when a disease will spread, how a chemical mixture can keep time, and why weather forecasts run out after about two weeks.
Models that explain, not just predict
The two previous topics built a toolkit: find the equilibria, linearise, look for conserved quantities, limit cycles and Lyapunov functions. Applied models are where that toolkit earns its keep. A good model is simple enough to analyse and states its assumptions plainly, so that when its predictions fail, you can see which assumption was to blame. This topic follows four families of models through ecology, epidemiology, chemistry and biology, and weather. In each, the qualitative behaviour, such as cycling, damping, a threshold or sensitivity, matters more than any formula for the solution.
Predator-prey cycles
The Lotka-Volterra model from the topic Modeling with systems, with prey \(x\) and predators \(y\), is \(x' = ax - bxy\), \(y' = -cy + dxy\), with \(a, b, c, d > 0\). Its equilibria are \((0, 0)\), a saddle, and the coexistence point \((\frac{c}{d}, \frac{a}{b})\). There the Jacobian has \(\tau = 0\) and \(\Delta = ac > 0\): a linear centre, the inconclusive case. A conserved quantity settles it:
satisfies \(H' = 0\) along solutions, and its level curves around the equilibrium are closed. So the populations cycle for ever. Over each cycle their average values equal the equilibrium values.
Worked example: a predator-prey cycle
Take \(a = 1\), \(b = 0.5\), \(c = 0.75\), \(d = 0.25\), with time in years and populations in hundreds. 1. Equilibrium: \((\frac{c}{d}, \frac{a}{b}) = (3, 2)\), so 300 prey and 200 predators. 2. The Jacobian, evaluated at \((3, 2)\), is
3. Its eigenvalues are \(\pm i\sqrt{0.75} \approx \pm 0.866i\), so small cycles have period about \(\frac{2\pi}{0.866} \approx 7.3\) years. 4. Starting from a prey peak, the predators peak about a quarter of a cycle later, roughly 1.8 years: more prey feeds more predators, which then cut the prey back. 5. Check by simulation: a cycle starting near \((3.3, 2)\) has period about 7.3 years and averages \((3, 2)\).
Worked example: prey that limit themselves
Now assume logistic prey growth with carrying capacity \(K = 6\) (600 prey): \(x' = x(1 - \frac{x}{6}) - 0.5xy\), with \(y' = -0.75y + 0.25xy\) as before. 1. Coexistence needs \(y' = 0\), so \(x = 3\); then \(x' = 0\) gives \(1 - 0.5 = 0.5y\), so \(y = 1\). 2. At \((3, 1)\), \(f_x = 1 - \frac{x}{3} - 0.5y = -0.5\), \(f_y = -0.5x = -1.5\), \(g_x = 0.25y = 0.25\), \(g_y = 0.25x - 0.75 = 0\). 3. So \(\tau = -0.5\) and \(\Delta = 0.375\), with \(\tau^2 < 4\Delta\): a stable spiral. 4. The cycles now damp towards 300 prey and 100 predators. One changed assumption turned neutral cycles into a steady state. Other refinements, such as predators that become satiated, can instead produce a stable limit cycle.
The SIR epidemic model
Split a population into fractions \(s\) susceptible, \(i\) infected and \(r\) recovered. Assume the population is closed and fixed over the outbreak (no births, deaths or migration), people mix uniformly, infected people recover at a constant rate \(\gamma\), and recovery gives lasting immunity. Then
where \(\beta\) is the transmission rate. Since \(i' = i(\beta s - \gamma)\), infections grow only while \(s > \frac{\gamma}{\beta}\). The basic reproduction number \(R_0 = \frac{\beta}{\gamma}\) is the average number of people one case infects in a fully susceptible population. An outbreak can start only if \(R_0s_0 > 1\), infections peak when \(s = \frac{1}{R_0}\), and herd immunity needs an immune fraction above \(1 - \frac{1}{R_0}\).
Worked example: an outbreak threshold
A disease has \(\beta = 0.5\) per day and \(\gamma = 0.2\) per day, an infectious period of about 5 days. 1. \(R_0 = \frac{0.5}{0.2} = 2.5\), and the rates have the same units, so \(R_0\) has none. 2. In a fully susceptible population \(R_0s_0 = 2.5 > 1\): an outbreak. 3. Infections peak when \(s\) falls to \(\frac{1}{2.5} = 0.4\). 4. Dividing the first two equations gives \(\frac{di}{ds} = -1 + \frac{1}{R_0s}\); integrating from \(s \approx 1\), \(i \approx 0\) gives \(i = 1 - s + \frac{1}{R_0}\ln s\). At \(s = 0.4\), the peak is \(0.6 + 0.4\ln 0.4 \approx 0.23\): about 23% infected at once. 5. Herd immunity needs more than \(1 - 0.4 = 60\%\) immune. If 70% are vaccinated first, \(R_0s_0 = 0.75 < 1\) and no outbreak can start.
Chemical clocks and nerve cells
Some chemical reactions oscillate on their own. In the Belousov-Zhabotinsky reaction, found by Boris Belousov in the 1950s and studied by Anatol Zhabotinsky in the 1960s, a stirred mixture switches colour back and forth for many minutes. Simplified models of it have a stable limit cycle born in a Hopf bifurcation, so a steady supply of reactants is enough to sustain a rhythm, with no periodic forcing. Nerve cells behave similarly. The FitzHugh-Nagumo model reduces a neuron to two equations: a small stimulus dies away, but one above a threshold triggers a large excursion, a spike, before the cell returns to rest. A steady strong stimulus produces a limit cycle of repeated firing.
Weather and chaos
In 1963 Edward Lorenz published a three-variable model of convection:
Here \(\beta\) is a geometric constant, not the SIR transmission rate. With \(\sigma = 10\), \(\rho = 28\) and \(\beta = \frac{8}{3}\), typical solutions never settle and never repeat, yet stay on a bounded, butterfly-shaped strange attractor. The system is deterministic, but two starts that differ very slightly separate roughly exponentially, so long-range prediction fails. This sensitive dependence on initial conditions is chaos. Meteorologists estimate the limit for detailed day-to-day weather forecasts at about two weeks, though climate statistics remain predictable. Chaos needs three or more dimensions in an autonomous system, as the Poincaré-Bendixson theorem showed.
Common mistakes to avoid
First, treating the neutral Lotka-Volterra cycles as robust: small changes to the model, such as prey self-limitation, destroy them. Second, confusing chaos with randomness; a chaotic system has no randomness in it at all. Third, computing \(R_0\) from rates in different units, such as \(\beta\) per day and \(\gamma\) per week. Fourth, using \(\frac{1}{R_0}\) as the herd immunity threshold: that is the susceptible fraction at which infections peak, and the immune fraction needed is \(1 - \frac{1}{R_0}\). Fifth, forgetting the assumptions, such as uniform mixing, when quoting a model's prediction.
Key takeaways
Lotka-Volterra predator-prey models have a linear centre at \((\frac{c}{d}, \frac{a}{b})\), and a conserved quantity makes the cycles closed, with predator peaks lagging prey peaks. Prey self-limitation makes the trace negative, so the cycles damp, as in the stable spiral of the example. The SIR model predicts an outbreak only when \(R_0s_0 > 1\), a peak when \(s = \frac{1}{R_0}\), and herd immunity above an immune fraction of \(1 - \frac{1}{R_0}\). Limit cycles explain chemical clocks and repeated nerve firing. The Lorenz equations show deterministic chaos, which limits detailed weather forecasts to about two weeks.
Flashcards
Flashcard
x′ = ax − bxy, y′ = −cy + dxy
Test your understanding
Question 1 of 5
For x′ = 2x − 0.1xy, y′ = −0.5y + 0.02xy, with x prey and y predators, where is the coexistence equilibrium?
Question 2 of 5
A disease has β = 0.3 per day and γ = 0.1 per day, and 75% of people are already immune. What does the SIR model predict for a newly introduced case?
Question 3 of 5
A disease has R₀ = 4. What fraction of the population must be immune for herd immunity?
Question 4 of 5
Logistic growth is added to the prey in a Lotka-Volterra model. What typically happens to the coexistence equilibrium?
Question 5 of 5
Which statement best describes the chaos in the Lorenz equations?
