The question this rung answers
By now you can solve plenty of homogeneous problems: a metal bar cooling with no heater inside it, a string vibrating with nothing pushing on it, a steady potential with no charge in the room. The tool was the superposition principle — add simple solutions, fit the data, done. But the real world is full of forcing: a heater warming the bar, a hand shaking the string, a charge density driving the field. That forcing shows up as an extra term with no unknown in it, the source term, turning a homogeneous equation into an inhomogeneous one like u_t - k u_xx = f(x, t).
Here is the question the whole rung answers, and it is wonderfully sharp: given an inhomogeneous linear PDE with any source f, can we write the solution down once and for all in terms of f? Not solve it again from scratch for each new heater, but build a single object — a machine — that takes any f as input and hands back the solution. The answer is yes, and the machine has a name we will reach by the end of this rung: a Green's function. This first guide builds the idea behind it, slowly and physically, before any formulas appear.
A sharp poke: the point source
Imagine the simplest forcing you can: not a heater spread along the whole bar, but a single, infinitely concentrated pulse — all of the heat dumped at one point x = 0 in one instant, nowhere and never else. As an idealization of "a unit of stuff packed into a single spot," this is the Dirac delta, written delta(x). It is zero everywhere except at one point, infinite there, and yet its total integral is exactly 1. No ordinary function can do that, which is the honest catch we will return to.
Why bother with such a violent, unphysical object? Because it is the perfect probe. A real source f(x) is spread out, complicated, different in every problem. But you can picture it as a dense crowd of tiny point sources, one standing at every location x', each carrying the local amount f(x'). If you knew how the system responds to one lone point source — the delta-source response — you would know how it responds to the whole crowd, simply by adding up the individual responses. The delta is the single brick from which every source is built.
The response, honestly: a distribution, not a function
Now feed that single poke into a linear operator L — say the heat operator, L u = u_t - k u_xx — and ask for the solution. The output is the response to a unit point source, and it gets its own name: the fundamental solution, the object E satisfying L E = delta. You have, in fact, already met a famous example without calling it this. The heat kernel, the bell-shaped Gaussian that spreads from a single hot spot, is the fundamental solution of the heat equation: it is literally the temperature you get from dropping one unit of heat at one point at one instant.
But here is the catch you were promised, and it must be told straight. Since delta is not a function, the equation L E = delta cannot be read in the ordinary, plug-in-a-point sense — and indeed the fundamental solution itself is often singular, blowing up where the source sits. The honest setting is the language of distributions, the rigorous framework you built earlier in this very ladder. There, delta is a perfectly respectable distribution, L E = delta is a precise statement (tested against smooth test functions), and E is a generalized function that may be infinite at a point yet still solve the equation distributionally. The intuition "response to a point source" is exactly right; the rigor is what makes "point source" mean something.
Stitching the pokes together: superposition as an integral
Now the payoff. Suppose L E = delta gives the response to a poke at the origin. By the symmetry of a constant-coefficient operator, a poke at any other point x' just produces a shifted copy E(x - x'). And a poke of strength f(x') produces f(x') E(x - x'), because the equation is linear: double the source, double the response. The real source f is the sum of all these pokes, so by superposition the solution is the sum — really the integral — of all their responses.
L E = delta (one unit poke at the origin -> the fundamental solution E) L u = f (any spread-out source f) u(x) = integral of E(x - x') f(x') dx' ( = E * f, a convolution ) Check: L u = integral of [L E](x - x') f(x') dx' = integral of delta(x - x') f(x') dx' = f(x)
That integral u = E * f is a convolution, and it is the whole magic in one line: you solved the equation once, for the single point source, and now every other source is handled by integrating. Build the machine once, run it forever. (Making sense of E * f when E is a distribution is, again, exactly what the earlier rung's convolution of distributions was for — the rigor is already in your hands.) The expression of a solution as such an integral against the source has a name we will lean on for the rest of the rung: a representation formula.
From free space to a room with walls
The fundamental solution E answers the point-source question in free space — an infinite bar, an unbounded plane, with no walls to obey. But most real problems live in a bounded region with boundary conditions: a rod of finite length held at fixed temperatures, a drum clamped at its rim, a charge inside a grounded metal box. There, the bare point-source response is not enough on its own, because it ignores the walls. The fix is to add a clever correction.
The full Green's function G for a boundary-value problem is built from two pieces: the free-space fundamental solution, which supplies the right singularity at the source, plus a smooth boundary corrector — an extra solution of the homogeneous equation, with no source of its own, tuned so that the sum satisfies the boundary condition exactly. The Green's function is that combined object: the response to a point source that also respects the walls. Guide 3 builds it carefully; guide 4 shows a beautiful trick, the method of images, for finding the corrector on simple geometries like half-spaces and balls.
The same idea, now in time
One last extension makes the point-source idea reach the time-dependent equations, the heat and wave equations. There, the natural poke is a source switched on at a single instant: a flash of heat at time s, or a single sharp tap on the string. The response to a poke at time s is felt only afterward, for t > s — causes precede effects — so the relevant kernel is built to vanish before the source fires. That respect for the arrow of time is exactly what a heat or wave problem demands, and it leads to the causal version of the Green's function.
When you superpose these timed pokes — integrating the response over all the instants s up to the present t — you get a familiar friend wearing a new costume. This is Duhamel's principle: solve the forced problem by summing the free evolutions of every instantaneous source. You likely met it for ODEs, and again inside the heat rung; now you can see why it is true. It is nothing but superposition of point-source responses, applied in the time direction. The full story — representation formulas in space, Duhamel in time — is the finale of this rung in guide 5.
So the plan for the rung is now in view, and it is all one idea seen from different angles. Find the response to a single point source — the fundamental solution (guide 2). Correct it to obey boundaries — the Green's function (guide 3). Build that correction cleverly with reflections — the method of images (guide 4). And assemble every answer by superposing the pokes, in space and in time — representation formulas and Duhamel's principle (guide 5). One sharp poke, fully understood, unlocks every forcing there is.