Diffusion that fights back
Everything you learned in the parabolic rung said diffusion is a peacemaker: the diffusion equation u_t = k u_xx takes any rough initial profile and smooths it relentlessly toward flatness, total stuff is conserved, and bumps only ever spread out and fade. Now add a single new term that does NOT conserve anything — a reaction that locally creates or destroys u depending on how much is already there. The result is the reaction-diffusion equation u_t = k u_xx + f(u): the u_xx part still shuttles material from crowded to empty places, but f(u) lets each point breed or die on its own. This is the simplest semilinear equation — linear in the highest derivatives, nonlinear only through the undifferentiated f(u) — and that one nonlinear term changes the whole character of the dynamics.
Think of u(x,t) as a population density spread along a line. Diffusion is the random wandering of individuals; the reaction f(u) is birth and death. A famous choice is logistic growth, f(u) = r u (1 - u): when u is small the population grows almost exponentially, but as u approaches the carrying capacity 1 the growth shuts off. Diffusion alone would let the population melt away; reaction alone would let it grow uniformly everywhere with no sense of place. Put them together and you get the genuinely new phenomenon this guide is about — an organized invasion that advances across empty territory at a sharp, well-defined speed.
A wave that carries its own shape
Here is the central object: a travelling wave. It is a solution that looks like one fixed profile sliding rigidly along the line at constant speed, u(x,t) = U(x - c t). The shape U never changes; the whole pattern just translates. We already met this idea for the wave equation, where any f(x - c t) travels — but there the speed c was handed to us by the equation. The astonishing thing about reaction-diffusion is that the equation does NOT tell you the speed; the wave has to discover c for itself, and only certain speeds are even possible.
To see why, substitute the ansatz. Write z = x - c t. Then u_t = -c U'(z) and u_xx = U''(z), so the whole PDE collapses into an ordinary differential equation in the single variable z: k U'' + c U' + f(U) = 0. The miracle is that the partial differential equation became an ODE — because the travelling-wave assumption removed time as an independent variable, freezing the picture in the frame that moves along with the front. We now look for a profile U(z) that climbs from one rest state to another: U going to 1 as z goes to minus infinity (territory already invaded) and U going to 0 as z goes to plus infinity (territory not yet reached).
PDE: u_t = k u_xx + f(u)
ansatz u(x,t) = U(z), z = x - c t
|
| u_t = -c U', u_xx = U''
v
ODE: k U'' + c U' + f(U) = 0
boundary conditions on the profile:
U(-infinity) = 1 (invaded state behind the front)
U(+infinity) = 0 (empty state ahead of the front)
--> only special speeds c admit such a connecting profileFor the logistic reaction f(u) = r u (1 - u), this is the celebrated Fisher-KPP equation, written down independently by Fisher (modelling an advantageous gene sweeping through a population) and by Kolmogorov, Petrovsky and Piskunov. A travelling front exists for every speed c at or above a minimum value c* = 2 sqrt(k r), and the front you actually see growing from a localized start always selects exactly that slowest admissible speed. So the invasion speed is set by just two numbers — how fast individuals diffuse and how fast they reproduce — combined as 2 sqrt(k r). This single formula predicts how fast a species spreads into new habitat, how fast a flame eats into fuel, how fast an advantageous gene takes over.
Bistable fronts: a tug-of-war with a definite winner
Fisher-KPP has one stable state (u = 1) and one unstable one (u = 0), so the invasion is one-directional: the full state always conquers the empty one. A richer and very different story comes from the Allen-Cahn equation, u_t = k u_xx + u (1 - u^2). Now there are TWO stable rest states, u = +1 and u = -1, separated by an unstable middle state u = 0. This is the classic model of two phases of a material — say two crystal orientations, or magnetization pointing up versus down — meeting along an interface. The travelling wave is now a front that connects -1 on one side to +1 on the other, and it moves so as to let the more-favoured phase eat the less-favoured one.
When the two phases are exactly equally favoured, the front sits still — c = 0 — and you can solve the profile in closed form: U(z) = tanh(z / sqrt(2k)), a smooth S-shaped step from -1 to +1 whose width is set by sqrt(k). Tip the balance slightly (make the reaction asymmetric, say f(u) = u(1-u^2) + a small constant) and the front starts to drift at a speed proportional to how lopsided the energies are. This is bistable front propagation, and it is the engine behind phase transitions, nerve-impulse fronts, and the coarsening of patterns over time, where larger domains slowly swallow smaller ones.
Turing's surprise: diffusion that builds patterns
Now for the result that genuinely overturns intuition. Everything so far still had diffusion playing its smoothing role; the reaction supplied the novelty. In 1952 Alan Turing asked a heretical question: could diffusion itself, the great equalizer, ever CREATE structure out of uniformity? With one chemical the answer is no. But take TWO reacting, diffusing chemicals — an activator u that promotes both itself and an inhibitor v that suppresses u — and couple them: u_t = D_u u_xx + f(u,v), v_t = D_v v_xx + g(u,v). Turing showed that a spatially uniform steady state, perfectly stable when the chemicals cannot move, can be DESTABILIZED the moment you let them diffuse. This is the Turing instability, and it is the most counterintuitive fact in this entire rung.
The mechanism has a vivid name: local activation, long-range inhibition. The key requirement is that the inhibitor diffuses much faster than the activator, D_v much larger than D_u. Picture a small accidental bump of activator. Locally it amplifies itself, so the bump wants to grow into a peak. But it also produces inhibitor, which races outward far faster than the activator spreads, suppressing activator in a wide ring around the peak. The result is a peak fenced off by a moat of suppression — and the natural spacing of those moats sets a preferred wavelength. The uniform state shatters into a regular array of spots or stripes whose size is chosen by the diffusion constants, not by the initial data.
How do you actually check this? Linearize around the uniform steady state and look at a small perturbation shaped like cos(q x) — a single spatial wavenumber q. Each wavenumber grows or decays at its own rate; the instability appears when some BAND of intermediate wavenumbers has a positive growth rate while q = 0 (the uniform mode) and very large q (fine wiggles, killed by diffusion) both decay. The fastest-growing q in that band is the wavelength you will see emerge. This is the same linear-stability-plus-wavenumber analysis you would run for any pattern-forming system, and it is the analytic heart of pattern formation.
What is true, what is a sketch, and where it lives
Be honest about what the linear analysis does and does not tell you. The Turing calculation only proves the uniform state is unstable and predicts the wavelength of the FIRST pattern to appear; it says nothing about the final amplitude or the exact shape, because once the perturbation grows the nonlinear terms take over and the linearization is no longer valid. Whether the system settles into spots, stripes, or hexagons, and how those compete and coarsen, is a fully nonlinear question that usually needs the computer or delicate weakly-nonlinear theory. Linear stability is a reliable detector of WHERE patterns begin, not a description of where they end up.
A second honesty: reaction-diffusion is a beautiful and predictive idealization, but it is a caricature of real biology and chemistry. The Belousov-Zhabotinsky reaction really does make travelling chemical waves and spirals in a dish, which is a genuine triumph. But whether the specific stripes on a fish or the spots on a leopard are literally produced by a two-chemical Turing mechanism is still debated — the equations capture the LOGIC of spontaneous patterning convincingly, while the actual molecular players are often unknown or more complicated. Use the model for insight into the mechanism, not as a claim that nature solves exactly these two equations.
Finally, place this guide in the rung. Fisher-KPP and Allen-Cahn fronts are nonlinear, but they remain SMOOTH and well-behaved forever — diffusion keeps the parabolic smoothing alive and nothing ever blows up. That gentleness is special to reactions that saturate. The very next guide asks what happens when the reaction does NOT saturate — when f(u) grows like u^p faster than diffusion can dissipate it — and there the solution can run away to infinity in finite time. So this guide is the calm before the storm: nonlinearity here organizes and structures, while the rest of the rung shows nonlinearity that destroys.