The one privilege we are about to lose
Look back at everything you have done so far on this ladder and ask what secretly powered all of it. The answer is a single sentence: if u and v both solve the equation, then so does a u + b v for any constants a and b. That is the superposition principle, and it is the privilege of being linear. It let you split a hard problem into easy pieces — single Fourier modes, single eigenfunctions, single point-source responses — solve each piece on its own, and then add the pieces back up to rebuild the answer.
Every triumph of the linear theory was really this trick wearing a different costume. Separation of variables is superposition over a discrete sum of modes; the Fourier transform is superposition over a continuous family of waves; a Green's function is superposition over a continuum of point sources, the solution written as an integral of responses to individual deltas. Behind each of them sits the same comfortable fact: the solutions of a linear PDE form a flat vector space, where adding and scaling never take you off the surface of valid answers.
A ladder of nonlinearity, from mild to wild
Nonlinearity is not one thing; it comes in grades, and naming the grade tells you how much trouble to expect. A semilinear equation keeps the top-order derivatives linear and stuffs all the nonlinearity into the lower-order term: the reaction-diffusion equation u_t = Laplacian u + f(u) is the model — diffusion is linear, the reaction f(u) is the troublemaker. A quasilinear equation lets the coefficients of the top derivatives depend on u and its lower derivatives, but those top derivatives still appear linearly, as in u_t + u u_x = 0.
At the far end sits the fully nonlinear equation, where even the highest derivatives are tangled up inside a nonlinear function — the eikonal equation |grad u| = 1 and the Hamilton-Jacobi equation u_t + H(grad u) = 0 are exactly this kind. The grading matters because it predicts the failure mode. Quasilinear first-order equations break by forming shocks (the coefficient u carries information at a speed that depends on u itself, so fast parts overtake slow ones). Semilinear reaction equations break by blowing up (the reaction f(u) feeds on its own output). Same loss of superposition, very different catastrophes.
linear L[a u + b v] = a L[u] + b L[v] heat, wave, Laplace semilinear u_t = Laplacian u + f(u) reaction-diffusion, u^p quasilinear a(u) u_x + b(u) u_t = c(u) Burgers, conservation laws fully nonlinear H(grad u) = 0 eikonal, Hamilton-Jacobi going down the list: more of u tangled into higher derivatives
What breaks: the toolbox stops working
When superposition dies, most of your hard-won machinery dies with it, and it is worth being honest about exactly how. You can no longer build a general solution out of a sum of modes, because adding two modes of a nonlinear equation does not give a solution — plug u + v into u u_x and you get cross terms u v_x + v u_x that have no right to be there. The Fourier transform, which turned derivatives into multiplication and decoupled the frequencies, now couples every frequency to every other one through the nonlinear term. A Green's function cannot exist in the old sense, because there is no linear operator to invert.
Even the words "the general solution" become slippery. Recall the honest fact from the very start of this ladder: a PDE's general solution involves arbitrary functions, not just arbitrary constants, so it is data — initial and boundary conditions — that pins down the answer. In the linear world the type of the equation (elliptic, parabolic, hyperbolic) told you which data was well-posed. Nonlinearity muddies even that: the same equation can change type as the solution evolves, and existence may hold for a while and then fail. We trade tidy formulas for a harder, more honest question — does a solution exist at all, and for how long?
What we gain: phenomena linearity could never show
Here is the surprise: losing superposition is not only a loss. A linear equation is in a sense too well-behaved — its solutions just sit in a flat vector space, scaling up and down in proportion, never producing anything qualitatively new. Nonlinearity is where richness lives, and the rest of this rung is a tour of four phenomena that simply cannot happen in a linear world. The first is finite-time blow-up: a solution that races to infinity in finite time. For the semilinear heat equation u_t = Laplacian u + u^p with p large enough, the reaction term feeds on its own value faster than diffusion can spread it out, and the solution literally ceases to exist after some finite time.
The second is pattern formation. A reaction-diffusion system, left to a uniform state that linear intuition says should be stable, can spontaneously break into stripes and spots — the Turing instability, where diffusion, normally a smoother, becomes a destabilizer when two species diffuse at different rates. The third is the soliton: a solitary wave that holds its shape forever and survives collisions unchanged. In the Korteweg-de Vries equation and the nonlinear Schrodinger equation, nonlinear steepening (which wants to form a shock) exactly balances dispersion (which wants to spread the wave out), and a stable lump of energy travels as a single coherent object.
The fourth is shock formation and non-uniqueness. You met this already in the hyperbolic rung: a smooth solution of a scalar conservation law can develop a genuine jump in finite time, after which the classical PDE is meaningless and we must pass to weak solutions. And the candid sting is that a weak solution is not unique — many functions satisfy the equation in the weak sense, and you need an extra admissibility rule, an entropy condition, to single out the physical one. Non-uniqueness, like blow-up, is a creature that linear theory simply does not have room for.
How to think when superposition is gone
Since you can no longer assemble answers from pieces, the strategy shifts from constructing solutions to controlling them. The central new tool is the energy method: instead of finding u, you track a single scalar quantity built from u — typically an integral of u^2 or of |grad u|^2 — and you differentiate it in time. If you can show that quantity stays bounded, the solution cannot run away, and you have proved existence for all time. If instead you can show it must reach infinity in finite time, you have proved blow-up. You give up the exact answer in exchange for an honest yes-or-no on survival.
Here is the honest frontier this rung opens onto. For many nonlinear PDEs the basic question — does a smooth solution exist for all time, or does it blow up? — is simply open. The most famous case is the three-dimensional Navier-Stokes equations: nobody knows whether a smooth flow can spontaneously form a singularity, and that question is a Clay Millennium Prize problem. So the right posture is humility plus a toolkit: linearize to test stability, build energy estimates to bound or trap the solution, and accept weak or viscosity solutions when classical ones cannot survive.
One last reframing to carry forward. The reason these four phenomena look so strange is that each is the result of two competing tendencies fighting to a draw — and the draw is something brand new. Steepening versus dispersion gives the soliton; reaction versus diffusion gives the Turing pattern and the travelling front; growth versus spreading gives blow-up or survival; compression versus the entropy condition gives the unique physical shock. Linear equations have no competing tendencies to balance, which is exactly why their world is so flat. The art of nonlinear PDEs is reading the balance, and that is what the rest of this rung will teach you to do.