From one solution to two
The previous guide handed you a one-sided fact about a single solution: by the weak maximum principle, a function with Laplacian u >= 0 inside a region (a subsolution) cannot have an interior maximum bigger than its boundary values, and the strong maximum principle sharpens that to say it cannot even touch its boundary maximum inside without being constant. That is already powerful, but notice it talks about one u at a time. The leap in this guide is small to state and enormous in consequence: point the very same principle not at u, but at the difference of two functions. Linearity does the rest.
Here is the engine in one breath. Suppose u and v are both harmonic in a region D, and suppose on the boundary u <= v everywhere. Form the difference w = u - v. Because Laplace's equation is linear, w is harmonic too (the superposition principle at work), and on the boundary w <= 0. The maximum principle says a harmonic w attains its maximum on the boundary, where it is at most 0 — so w <= 0 everywhere inside. That is exactly u <= v throughout D. We compared two solutions over the whole region using only their boundary ordering, and we never wrote down either one.
The comparison principle, stated and felt
Cleaned up, the engine is the comparison principle: if u is a subsolution and v a supersolution of the same linear equation, and u <= v on the boundary, then u <= v in the entire interior. Read it as a promise about order. Boundary order is preserved into the bulk — solutions cannot cross or overtake each other inside without first doing so on the rim. Picture two soap films stretched over the same wire frame, one rim held everywhere at or below the other: the lower film simply cannot bulge up through the higher one anywhere in between. The films are forbidden to intersect except possibly where their rims already meet.
One immediate dividend is positivity preservation, and it is just comparison with the constant zero. Take v = 0, which is trivially harmonic. If u is harmonic with u >= 0 on the boundary, then comparison gives u >= 0 throughout. In words: a steady temperature that is nowhere negative on the rim is nowhere negative inside. You cannot conjure a cold spot below freezing in the interior of a plate whose entire edge is held at or above freezing. The same statement holds for the heat equation: nonnegative initial and boundary temperatures stay nonnegative for all time — diffusion never invents negative heat.
COMPARISON PRINCIPLE (linear equation L)
L u >= 0 (subsol) u <= v on boundary
L v <= 0 (supersol) =====================> u <= v inside
special cases
v = 0 : u >= 0 on boundary ==> u >= 0 inside (positivity)
u, v both solutions, u = v on boundary ==> u = v inside (uniqueness)
|u| <= M on boundary ==> |u| <= M inside (a-priori bound)Squeezing out uniqueness
Now apply comparison in both directions at once and uniqueness falls out, almost for free. Suppose u and v solve the same boundary-value problem — same equation inside, same data g on the boundary. Then u <= v on the boundary (in fact u = v there), so comparison gives u <= v inside. But the situation is perfectly symmetric: swap the roles and v <= u inside as well. Two inequalities, u <= v and v <= u, can only both hold if u = v. The solution is unique. No formula, no integral, no separation of variables — just the order-preserving squeeze applied from above and below.
It is worth being honest about exactly what this buys and what it does not. Comparison delivers uniqueness and continuous dependence, but it says nothing about existence. It is entirely possible for a problem to have at most one solution and yet have none at all — the comparison argument would run vacuously, comparing two things that do not exist. Existence is a separate promise, earned by actually constructing a solution (the Poisson formula on a disk, separation of variables, the Lax-Milgram theorem in the weak theory). The qualitative tools of this rung are razor-sharp at constraining and distinguishing solutions; they are silent on conjuring one into being.
Adding time: the parabolic version
Everything above transplants cleanly to evolution in time once you know where the boundary really is. For the heat equation u_t = k u_xx the right setting is a space-time box, and the parabolic maximum principle says the maximum is attained on the parabolic boundary — the bottom (the initial data) and the two sides (the spatial boundary), but crucially not the top. Heat the rim and the start however you like; the interior at later times can never exceed the hottest value seen on that lower-and-sides frame. The future is hemmed in by the past and the edges, and it cannot run hotter than both.
With the parabolic boundary in hand, the comparison and uniqueness arguments run verbatim. Two solutions of the heat equation that agree on the initial line and on the spatial boundary must agree everywhere afterward: subtract, get a heat solution that is zero on the whole parabolic boundary, and the parabolic maximum principle squeezes it to zero for all later time. So the forward heat problem — initial temperature plus boundary temperature — has a unique solution. And comparison gives the same monotone ordering: warmer start and warmer rim means a warmer plate forever, instant by instant.
Why exclude the top? Because the arrow of time only points one way for diffusion. The principle controls the present using the past and the sides, never the future — exactly mirroring that the heat equation smooths going forward and is hopelessly ill-posed going backward. Trying to run it in reverse, the backward heat equation, is the irreversibility of diffusion made precise: tiny errors in a present temperature explode when you try to reconstruct the past. The parabolic boundary's missing top face is the maximum principle quietly encoding that one-way street.
Where the engine stops: the wave equation
Now the honest boundary of the whole method. The comparison machine runs on elliptic equations (Laplace) and parabolic ones (heat), because both have a maximum principle. The wave equation u_tt = c^2 u_xx has no maximum principle at all, and it is instructive to see why this is so rather than just be told. A travelling wave u(x,t) = f(x - c t) carries its profile along unchanged; a sharp pulse stays just as sharp. Two waves can be ordered at the start and yet cross each other later as their bumps slide past — nothing forbids overtaking, because nothing smooths the bumps out. The whole order-preserving picture simply does not apply.
This is not a defect of our tools but a true difference in physics, and it splits the qualitative theory into two camps. Diffusion has infinite propagation speed and the smoothing property: a bump instantly nudges everywhere and instantly rounds off, and that smoothing is precisely what underwrites a maximum principle. Waves have finite speed c and instead exhibit propagation of singularities: a kink in the initial data does not heal — it travels along the characteristics forever, riding at speed c. Where heat blurs, waves carry. Guide 4 of this rung is devoted to exactly this divide; here it explains the precise reach of the comparison engine you have just built.
Two parting threads tie this guide to what comes next. First, comparison is the floor, not the ceiling: the strong maximum principle makes the ordering strict (if two solutions are ever equal at an interior point without being identical, something special must hold), and Harnack's inequality — guide 3 — pushes further still, bounding how large a positive solution can be at one interior point by its value at another, so positive solutions cannot vary too wildly. Second, even without a maximum principle the wave equation is not lawless: it has its own conserved quantity, the wave energy, and energy methods give its uniqueness — the subject that opens the final guide of this rung.