From "no interior peak" to "no interior lopsidedness"
You arrived here with two tools already sharp. The first guide gave you the strong maximum principle: a non-constant harmonic function attains its maximum and minimum only on the boundary, never in the open interior. The second gave you comparison and the uniqueness it buys. Harnack's inequality is the next floor up, and it is a genuine surprise. The maximum principle is a statement about one number — the peak — and where it can sit. Harnack is a statement about the whole interior at once: it says a positive solution cannot be tall in one room and nearly flat in the next. Its high values and its low values, measured over any region safely inside, are chained together.
Here is the picture in words. Take a positive harmonic function u — one obeying Laplacian u = 0 — on some big region, and look only at a smaller region K sitting comfortably inside, not touching the boundary. Harnack's inequality says there is a single constant C, depending only on the shapes of K and the big region (and the dimension), such that the largest value of u on K is at most C times the smallest value of u on K. In symbols, max over K of u is at most C times min over K of u. The constant does not care which particular harmonic function you chose, nor how big or small u is overall — scale u up by ten and both sides scale by ten, so the ratio is untouched. It is a universal speed limit on how unequal a positive solution can be inside.
Where the estimate comes from
You can almost see it through the mean-value property you met on Laplace's equation. Recall that a harmonic function equals the average of itself over any ball: u at the centre is the average of u over the surrounding sphere. Now take two interior points p and q that are not too far apart. Draw a small ball around p and a small ball around q; because p and q are close and both balls live inside the region, you can fit a slightly bigger ball centred at q that contains the small ball around p. The average of a positive function over a big ball is at least a fixed fraction of its average over any smaller ball sitting inside — the big ball only adds more positive stuff. Chase the averages and you get u(p) bounded by a constant times u(q). Connect any two interior points by a short chain of such overlapping balls, multiply the constants, and you have Harnack.
That chaining picture explains the two hypotheses at once. The constant grows with the number of balls needed to walk from one point to another, so points deep inside a fat region are cheaply connected (small C), but points hugging opposite sides of a thin sliver, or creeping toward the boundary, need a long expensive chain (huge C). And positivity is what let the averaging argument run — "a bigger ball has at least as much average" is only true when the integrand never goes negative. Lose the sign and the bookkeeping collapses.
Region D (harmonic, u > 0)
+-------------------------------+
| |
| [ K = compact inside ] |
| .-o---o---o---o-. | walk p -> q by overlapping balls
| p q | max_K u <= C * min_K u
| '-----------------' | C depends on K, D, dimension -- NOT on u
| |
+-------------------------------+
boundary (C -> infinity as K approaches it)What it forces a solution to be
The reason this dry-looking ratio is one of the load-bearing walls of the whole subject is that it has teeth. First, it sharpens the maximum principle into the strong form: if a positive harmonic function ever touches zero at an interior point, its minimum over a neighbourhood is zero, so Harnack forces the maximum there to be at most C times zero — that is, zero everywhere nearby. A positive harmonic function that dips to zero inside is therefore identically zero. That is the strong maximum principle, falling straight out of an inequality about averages.
Second, it gives you Liouville's theorem almost for free. Suppose u is harmonic and positive on all of space. Take any two points and surround them by an enormous ball; Harnack ties their values by a constant that, as the ball grows to fill the plane, settles to a fixed number independent of how far apart the points are. A function whose values everywhere are pinned within a fixed ratio of one another, no matter the distance, has nowhere to go: it must be constant. So a positive harmonic function on the whole plane is constant — a rigidity that looks like magic until you see Harnack underneath it.
Third, and this is the deepest payoff, Harnack-type inequalities are the engine of interior regularity. From a one-sided ratio bound you can extract an oscillation bound, then a Hölder estimate on the solution, then on its derivatives — a bootstrapping ladder that ends with the solution being smooth inside no matter how rough its boundary data was. This is exactly the road the De Giorgi-Nash-Moser theorem walks: it proves a Harnack inequality even for divergence-form equations whose coefficients are merely bounded and measurable — no continuity assumed — and from that single estimate it harvests Hölder continuity of every solution. That theorem cracked open Hilbert's nineteenth problem, and Harnack is its beating heart.
The parabolic version: a waiting time appears
Everything so far was elliptic — Laplace's equation, no time. The same idea holds for the heat equation u_t = k u_xx (or with the full Laplacian, u_t = k Laplacian u), and the way it changes is one of the most physically honest facts in PDE. For a positive solution of the heat equation, the value at one point controls the value at another point — but only if you give the heat time to arrive. The parabolic Harnack inequality compares u at a point (x, t) with u at a later point (y, s) with s strictly greater than t. The future is controlled by the past, with a constant depending on how far apart x and y are and how much time has elapsed.
That mandatory time lag is not a technical blemish — it is the physics of diffusion. A hot spot at one place takes time to make itself felt elsewhere; you cannot bound the temperature at a far point at the same instant, because heat has not reached it yet. And the lag points only forward, which is the inequality's way of encoding the irreversibility you learned on the heat equation: the heat equation runs cleanly forward in time and is hopelessly ill-posed backward. Harnack respects that arrow. Comparing later-to-earlier works; comparing earlier-to-later at the same time does not.
Harnack and the smoothing of diffusion
The parabolic Harnack inequality is the precise, quantitative face of the smoothing effect you first met as a slogan: the heat equation instantly irons out roughness. A solution whose interior values are all pinned within a fixed ratio of one another simply cannot have a sharp local feature — no spike, no crease, no sudden jump survives, because such a feature would make the local max far exceed the local min and break the Harnack ratio. Run the heat equation for any positive time and Harnack guarantees the solution is not just continuous but as smooth as you like inside. The estimate and the smoothing are two views of one fact.
This is also where Harnack quietly draws a border that the next guide will make its whole subject. The inequality is a creature of the elliptic and parabolic world — equations that average, equilibrate, and diffuse. It has no counterpart for the wave equation u_tt = c^2 u_xx, and the reason is not a missing proof but a wall of physics. Waves carry singularities along characteristics at finite speed and do not smooth: a kink in the initial data rides forever and stays a kink. There is no positive ratio bound to be had, because a wave can be large on a crest and exactly zero in the trough a wavelength away, with nothing in between forcing them together. Harnack is precisely the gift that diffusion gives and hyperbolic propagation withholds.