A solution that runs out of time
By now this rung has trained one reflex hard: in a nonlinear PDE you cannot add solutions, so the old comfort of the superposition principle is gone, and brand-new behaviour is allowed in. The previous guide showed one face of it — a reaction-diffusion equation organising itself into a travelling front or a Turing pattern. This guide shows the opposite face, the violent one. Sometimes a perfectly smooth, perfectly innocent initial state grows without bound and reaches infinity at a finite clock time. After that instant there is simply no solution to speak of. We call this finite-time blow-up.
The cleanest place to meet it is the semilinear heat equation u_t = Laplacian u + u^p, with p > 1 and a non-negative initial temperature. Read it as a tug of war. The Laplacian is the diffusion you already know from the heat equation: it spreads heat out, flattens peaks, and on its own would calm everything down. The term u^p is a source — a reaction that manufactures more u, and manufactures it faster and faster as u grows, because the exponent p is bigger than 1. The whole drama of this guide is which side wins: the calming diffusion, or the self-amplifying reaction.
The tiny ODE that predicts the catastrophe
To see why finite-time really can mean finite-time, throw away diffusion for a moment and keep only the reaction. Imagine the temperature is the same everywhere in space, so Laplacian u = 0 and nothing diffuses. Then u no longer depends on position, and the PDE collapses into an ordinary differential equation you can solve by hand: y'(t) = y^p. This is the pure reaction with no cooling — the worst case for the heat, the best case for catastrophe — and it is small enough to integrate in one line.
y'(t) = y^p , y(0) = a > 0 , p > 1 Separate: y^(-p) dy = dt Integrate: y^(1-p) / (1 - p) = t + C Use y(0)=a: C = a^(1-p) / (1 - p) Solve: y(t) = [ a^(1-p) - (p-1) t ]^(-1/(p-1)) Denominator hits 0 when t = T* = a^(1-p) / (p - 1) => y(t) -> +infinity as t -> T* (BLOW-UP at a finite time T*) Contrast p = 1: y' = y => y = a e^t (finite for every finite t)
Look hard at the blow-up time T* = a^(1-p) / (p - 1). Three honest lessons fall straight out. First, blow-up is real and the time is explicit — the denominator of y hits zero at T*, and y goes to infinity, no hand-waving. Second, hotter starts blow up sooner: a larger initial value a makes a^(1-p) smaller (since 1 - p < 0), so T* shrinks. Third, the formula is meaningless past T*; the solution does not 'come back down' — it ceases to exist, which is why we must speak of a maximal existence time and stop there. This little ODE is the heartbeat of the whole phenomenon.
Making it rigorous: trapping the PDE below an exploding ODE
A skeptic should object at once: you deleted diffusion to get that ODE, but real diffusion is still running, and diffusion cools. Could the Laplacian bleed off enough heat to rescue the solution? For a large enough hot blob the answer is no, and the tool that proves it is the comparison principle you met in the parabolic theory. Its slogan is simple and powerful: for these equations, if one solution starts below another and obeys the same kind of inequality, it stays below for all time. Order at the start is preserved in time. We will use it to sandwich the true PDE solution above a quantity we already know explodes.
- Pick a candidate lower barrier. Take the spatially-constant ODE solution y(t) from the last section, started at a = the minimum of the initial data. It solves y' = y^p exactly, with no diffusion term.
- Check it is a subsolution. Plug y into the PDE operator: y_t - Laplacian y - y^p = y^p - 0 - y^p = 0, and since y is constant in space the diffusion term vanishes. So y satisfies the equation with equality — it is an honest subsolution of u_t = Laplacian u + u^p.
- Order the starting data. At t = 0 we chose a to be the smallest value the real initial temperature ever takes, so u(x,0) >= y(0) everywhere in space. The true solution starts on top of the barrier.
- Apply the comparison principle. Because u starts at or above the subsolution y and both obey the same reaction-diffusion inequality, u(x,t) >= y(t) for as long as u exists. The true solution is pinned above the barrier.
- Read off the conclusion. y(t) blows up at the finite time T*, and u sits above it, so u must blow up no later than T*. Diffusion did not save the solution; it merely cannot push u below an exploding floor.
That is the whole proof, and its elegance is worth pausing on. We never solved the PDE; we only built a humble object we could solve by hand and showed the real solution could not escape underneath it. This is the spirit of a a priori estimate — extracting a hard fact about a solution (here, that it must die by time T*) before, and without, actually constructing it. Notice the asymmetry too: the argument needs the initial blob to be hot enough that even its minimum reaction wins. Small, cool data may diffuse away to nothing instead. Whether you blow up or fade out can hinge on the size of what you start with.
The critical exponent: when can diffusion win?
The comparison argument shows large data blows up, but it leaves a sharper question open: for small data, can diffusion ever win and let the solution live forever? The astonishing answer, found by Fujita, is that it depends on a single number — a critical exponent. For u_t = Laplacian u + u^p on all of n-dimensional space, the dividing line is p* = 1 + 2/n. The exponent p is compared against this threshold, and the verdict flips cleanly as you cross it.
Here is what the threshold says, in words. For 1 < p <= p* — the subcritical case — EVERY non-trivial non-negative solution blows up in finite time, no matter how small or gentle the data. For p > p* — supercritical — small enough data can exist globally and live forever, while large data can still blow up. Plug in the dimension and the number is concrete: p* = 3 in one dimension (n = 1), p* = 2 in the plane (n = 2), and p* = 5/3 in three dimensions (n = 3). Notice that as n grows the threshold drops toward 1, so higher dimensions make global survival easier — diffusion has more directions to spread into.
Sit with how strange that subcritical regime is. When 1 < p <= p*, there is no safe initial state at all: take data as tiny and gentle as you like, and the solution still finds a way to blow up in finite time. Diffusion is simply too weak in low dimensions to drain heat faster than even a faint reaction manufactures it. Above the critical exponent the balance tips: now diffusion, with more room to spread in higher dimensions, can win against small data, and global-in-time solutions exist — though a big enough hot blob will still explode by the comparison argument above. The exponent is the exact knife-edge between those two worlds.
One honest caveat about the critical case itself. The boundary p = p* belongs to the blow-up side here — at the exact threshold, even small data still blows up — but that endpoint is delicate and was settled only after Fujita's original work. The broader lesson, true across nonlinear PDE, is that a critical exponent is rarely a polite cutoff: the borderline case usually demands its own careful argument and often behaves like the side you would least expect. When you meet such a threshold, treat the equality case as a separate question, not a footnote.
What blow-up actually looks like — and what we still don't know
It helps to picture the explosion in space, not just in time. Often the solution does not blow up everywhere at once; it concentrates. A spike grows over one region — a blow-up point — while the rest of the field stays bounded, and near that point the profile frequently organises itself into a self-similar shape: the same bump, rescaled in height and width by powers of the remaining time T* - t, repeating itself as it sharpens toward the singular instant. This is a recurring trick of nonlinear PDE — near a singularity, zoom in with the right rescaling and a clean, universal profile emerges from the chaos.
Step back and notice that we have now seen two completely different ways a smooth solution can break down in finite time, and it is worth keeping them apart. In the first-order conservation laws earlier on this ladder, the value u stayed bounded but characteristics crossed and the slope went infinite — that is gradient blow-up, the birth of a shock, fixed by passing to a weak solution. Here, in the semilinear heat equation, the value u itself races to infinity. These are genuinely different catastrophes: one is a steep wall the solution survives in a weaker sense, the other is an amplitude that escapes the number line entirely.
End on an honest note about the edge of knowledge. For this clean model equation the blow-up story is largely understood, but the moment you complicate it — different reaction terms, systems, other geometries — telling global existence from blow-up turns hard fast, and for many important nonlinear PDEs it remains genuinely open. The headline example sits one rung over: whether smooth solutions of the three-dimensional Navier-Stokes equations can blow up in finite time, or live forever, is a famous unsolved problem. The simple ODE-and-comparison machinery you just learned is the first thing experts still reach for, precisely because the general question has no general answer.