Deterministic models

Deterministic models are systems whose future state is fully determined by the current state and the rules. Given the same starting point, they always produce the same trajectory. It is tempting to equate deterministic with predictable, but even the simplest non-linear rule can generate behaviour that is, for all practical purposes, indistinguishable from randomness.

A warm-up: the logistic map

Before we meet the SIR and its siblings, it is worth spending a moment on a one-line model that already contains most of the surprises deterministic dynamics has to offer. The logistic map is

\[x_{t+1} = r\, x_t\, (1 - x_t)\]

where \(x_t \in [0, 1]\) is a population at generation \(t\) and \(r\) is a growth rate. It was popularised by Robert May (may1976?) as a cautionary tale: a deterministic rule this simple can produce stable equilibria, clean periodic cycles, sudden bifurcations, and full-blown deterministic chaos — all by turning a single knob.

Interactive demonstration

Drag the r and x₀ sliders (or click anywhere on the bifurcation diagram) to set the dynamics.

  • The time series is your trajectory \(x_0, x_1, x_2, \dots\). Drag along it to pick which step the formula explains — the selected step and its predecessor are marked. Horizontal dashed lines show the attractor values where the sequence eventually settles.
  • The formula panel shows the rule with the numbers from that step plugged in. Hover any value in the computation to see the matching point highlighted on the time series.
  • The bifurcation diagram plots the attractor for every \(r\). The dots on the vertical marker are exactly the same values drawn as dashed lines in the time series above — same \(r\), same values, same colour.

A few landmarks worth hunting for with the sliders:

  • \(r \le 3\): a single stable fixed point at \(x^{*} = 1 - 1/r\). The time series settles on one dashed line.
  • \(r \approx 3\): the first bifurcation. The single line splits into two.
  • \(r \approx 3.449\), \(3.544\), \(3.564\), …: period-doubling cascade — 4, 8, 16, … values appear, with the ratio between successive bifurcations converging to the Feigenbaum constant \(\delta \approx 4.6692\).
  • \(r \approx 3.5699\): end of the cascade, onset of chaos. The list of attractor values collapses into a continuous band and the trajectory no longer repeats.
  • \(r \approx 3.828\): a conspicuous period-3 window in the middle of chaos — a reminder that the chaotic regime is not uniformly chaotic.
  • \(r = 4\): the dynamics are conjugate to a fair coin toss on a binary expansion; purely deterministic, yet sampling from it is statistically indistinguishable from randomness.

The moral, and the reason this detour belongs at the start of a deterministic modelling book, is that “deterministic” does not imply “tame”. When we turn to epidemic models in the next chapters — where coupling, thresholds, and time-varying parameters quickly outpace the logistic map in complexity — the intuition we have just built is the right one to carry in.