shock formation
A smooth ocean swell can steepen until its front goes vertical and it breaks into white water. Sound from an explosion sharpens into a bang. In both cases something perfectly smooth becomes, in finite time, a near-discontinuity. Shock formation is the mathematical name for this: a nonlinear conservation law turning smooth initial data into a genuine jump, all on its own.
Here is the mechanism, for u_t + f(u)_x = 0. Each value of u travels along a straight characteristic at its own speed f'(u). If higher values move faster than lower values just ahead of them — that is, if the speed f'(u) is decreasing in the direction of motion — then fast characteristics overtake slow ones. Where two characteristics carrying different values of u meet, the solution would have to take two values at once: impossible for a smooth function. So at that first crossing time the spatial derivative u_x becomes infinite (a gradient blow-up) and a shock is born. For Burgers' equation u_t + u u_x = 0 with decreasing initial data u_0, the first crossing happens at time t* = -1 / min(u_0'), explicitly.
Why it matters: shock formation is THE reason classical solutions are not enough for nonlinear hyperbolic PDEs. No matter how smooth and gentle your data, a shock can appear in finite time, after which u_t + f(u)_x = 0 cannot hold pointwise. This forces the entire apparatus of weak solutions, jump conditions, and entropy conditions. It is not a numerical artifact or a modelling flaw — it is a true feature of the equation, the reason sonic booms and hydraulic jumps exist.
Take Burgers' equation u_t + u u_x = 0 with initial data u_0(x) = 1 for x < 0 decreasing to 0 for x > 1. The taller parts (speed 1) chase the shorter parts ahead (speed 0). The profile steepens, and at the first time characteristics cross the slope goes vertical: a shock has formed, and from then on u jumps from 1 down to 0 across a moving front.
Compressive (decreasing) data forces characteristics to cross and a shock to form in finite time.
A common misconception is that shocks come from rough or noisy data. They do not: perfectly analytic initial data forms shocks, purely from the nonlinearity. Conversely, expansive (increasing) data spreads into a smooth rarefaction and never shocks — the direction of the gradient is what decides.