The promise: read off the extremes without solving
You arrive at this rung already able to solve a great many equations — separation of variables, the mean-value property, Green's functions, transforms. This rung asks a different and quietly powerful question: what can you say about a solution without solving it at all? The first and most useful answer is the maximum principle. It is a statement about where a solution's largest and smallest values are allowed to sit, and it costs you nothing but the equation itself — no formula, no series, no integral.
Here is the picture in one sentence. Take a steady temperature in a metal plate with no heater inside — a harmonic function solving Laplacian u = 0. The hottest and the coldest points are always on the edge of the plate, never strictly inside. That feels physically obvious: with no internal source, heat cannot pile up into a hidden interior hot-spot; any peak would immediately leak heat outward and flatten. The maximum principle turns that intuition into a theorem, and then — this is the surprise — extends it far beyond harmonic functions to whole families of elliptic and parabolic equations.
The weak maximum principle: extremes live on the boundary
Start with the cleanest case. Let u be harmonic in a bounded region D and continuous up to the wall. The weak maximum principle says: the maximum of u over the closed region (interior plus boundary) is attained on the boundary. In symbols, max over closure of D of u equals max over boundary of D of u. The word weak does not mean feeble — it means the theorem allows the maximum to also occur somewhere inside; it only refuses to let the inside beat the boundary. Apply the same statement to -u and you get the matching fact for the minimum: the coldest point is on the boundary too.
The proof idea is a lovely one-line trick worth carrying with you. Suppose first that u were strictly subharmonic, Laplacian u > 0 everywhere. At any interior maximum the function must curve downward in every direction, so its second derivatives u_xx and u_yy are both less than or equal to zero, forcing Laplacian u = u_xx + u_yy to be less than or equal to zero — flatly contradicting Laplacian u > 0. So a strict subharmonic function simply cannot have an interior max; the maximum is squeezed onto the boundary. To handle the borderline harmonic case Laplacian u = 0, nudge it: look at u + epsilon|x|^2, which has Laplacian equal to a small positive constant, apply the strict result, and let epsilon shrink to zero. That epsilon-nudge is one of the most reused moves in the whole subject.
Notice what the proof actually used: only that the second derivatives have a definite sign at a peak. It never used a formula for u. So the very same argument runs for any operator of the form a u_xx + b u_xy + c u_yy plus lower-order terms, as long as the second-order part is elliptic — meaning the discriminant condition b^2 - a c < 0 that you met when classifying second-order equations, the condition guaranteeing the operator curves consistently in all directions. This is why the maximum principle is the signature property of the elliptic and parabolic worlds and is absent for the hyperbolic wave equation, whose solutions happily oscillate to interior peaks.
The strong maximum principle: no interior tie, ever
The weak principle leaves a door ajar: it allows an interior point to equal the boundary maximum, as long as it does not exceed it. The strong maximum principle slams that door shut. It says: if a harmonic (or subharmonic) function attains its maximum at even a single interior point of a connected region, then u is constant throughout the entire region. There is no in-between. Either the maximum is strictly confined to the boundary, or the function is flat everywhere. A genuine, non-constant solution can never tie its own boundary record from the inside.
Where does this rigidity come from? Straight from the mean-value property you proved in the Laplace rung. A harmonic function equals its own average over every small sphere centred at a point. Suppose u hits its global maximum M at an interior point p. Then M = u(p) = average of u over a tiny sphere around p. But every value on that sphere is at most M. An average of numbers that are all at most M can only equal M if every one of them is exactly M. So u equals M on a whole little ball around p — and now repeat the argument from each newly-flat point, spreading the value M outward. Since the region is connected, the flat patch grows until it fills everything. That is the strong principle in a single breath: a peak that ties its average has nowhere to hide, and the flatness is contagious.
The Hopf lemma: even the boundary slope tells a story
The strong principle leaves one last refinement, and it is exactly the tool the rest of qualitative theory leans on. Suppose a non-constant harmonic function does attain its maximum, necessarily at a boundary point p (we just ruled out the interior). The Hopf boundary point lemma says something sharp about how it gets there: the outward normal derivative at p is strictly positive — u is strictly increasing as you walk out toward p. The function cannot flatten out and kiss its maximum tangentially at the wall; it must arrive with genuine, non-zero slope.
Why care about a slope at the wall? Because it is exactly what you need to make the principle work for the Neumann problem, where the boundary data is a slope rather than a value. If the normal derivative could vanish at a maximum, you could never conclude anything from Neumann conditions; Hopf's strict positivity is what rescues uniqueness there. It is also the technical heart of the moving plane method, which uses these boundary-slope facts to prove that solutions of many symmetric problems must themselves be symmetric — a striking consequence pulled, again, from the maximum principle and nothing more.
Time joins in: the parabolic maximum principle
Everything so far was elliptic — equilibrium, no time. But the same spirit governs diffusion. For the heat equation u_t = k u_xx, the parabolic maximum principle says the hottest point of a heated bar over a time interval occurs either at the start (the initial data) or on the spatial boundary (the ends of the bar) — never freshly born at an interior point at a later time. Once again it is physically transparent: with no internal heater, heat only flows from hot to cold, so a new interior hot-spot cannot appear out of nowhere after the clock starts.
Parabolic cylinder over a bar 0 <= x <= L, time 0 <= t <= T:
t = T +-----------------------+ <- TOP: not part of the parabolic boundary
| |
t | interior | the max of u is attained on the
| (no max born here) | shaded parabolic boundary below,
| | never strictly inside or on the top
t = 0 +#######################+
x=0 initial data t=0 x=L
#########################
parabolic boundary = bottom (t=0) + the two sides (x=0 and x=L)The asymmetry of the box is the whole point, and it is worth pausing on. The maximum can be reached at the final time t = T, but it is never strictly inside the cylinder for the first time — it had to arrive from the bottom or the sides. This lopsided box is the mathematical fingerprint of time's one-way arrow: diffusion only smooths going forward, which is precisely why running the heat equation backward is ill-posed. The strong parabolic version then mirrors the elliptic one: if a non-constant solution ever ties its earlier maximum at an interior point, the entire past — every earlier time — must have been constant, an even more dramatic rigidity than the elliptic case.
Honest fine print, and where it leads
Be honest about the conditions, because the principle fails the moment they break. First, you need the right sign structure: the bare equation Laplacian u = 0 obeys it, and so does Laplacian u + (lower-order terms) u_x... but a term like +c(x) u with a positive coefficient c can manufacture interior peaks and destroy the principle — this is why the equation -Laplacian u + c u with c >= 0, where the zeroth-order term has the helpful sign, is the standard safe setting. Second, you genuinely need ellipticity (or its parabolic analogue): drop it and the wave equation is the counterexample, with interior maxima all over the place. Third, the region must be bounded for the clean boundary statement; on unbounded domains a solution can sneak its supremum off to infinity unless you add a growth condition.
One more caution against over-reading the principle. It controls the values of u, not its derivatives in general — knowing the max is on the boundary tells you nothing directly about how steep or wiggly u is inside, except at an actual extremum via Hopf. And it is fundamentally a one-sided, inequality-flavoured tool: it loves Laplacian u >= 0 and Laplacian u <= 0, which is why it pairs so naturally with sub- and super-solutions rather than with exact formulas. Used within its limits it is almost magical; pushed past them it quietly lies.
Where does all this lead? Straight into the next four guides. Turn the principle into a tool for ranking two solutions and you get the comparison principle and the uniqueness it hands you for free (guide 2). Push the strong principle quantitatively — by how much must a positive solution at one point control its value at another? — and you arrive at Harnack's inequality (guide 3). Notice that the principle is intimately tied to smoothing, and contrast it with hyperbolic equations that refuse to smooth (guide 4). The maximum principle is not one theorem among many; it is the spine running down the whole back of qualitative PDE theory.