A hot bar and the question it asks
You arrive at this rung already fluent: you know an unknown u(x,t) depends on space and time, you have carried profiles along characteristics, and you have classified second-order equations into elliptic, parabolic, and hyperbolic. Now meet the head of the parabolic family. Picture a thin metal bar, lagged so no heat escapes through its sides, with one end warmed and the rest cool. Heat creeps along it; the temperature pattern softens and spreads as the minutes pass. The function u(x,t) is the temperature at position x along the bar at time t, and our whole task is to find the single equation that says how u changes from one instant to the next.
The end result is famous and short: u_t = k u_xx, the heat equation, where k is a positive constant. But a name learned without its origin is a hollow thing. The honest pleasure here is that this equation is not handed down from on high — it is forced on us by two facts so plain you already believe them. First, heat flows from hot to cold. Second, heat is not created or destroyed inside the bar; it only moves. Turn those two sentences into algebra and the equation simply falls out. The same derivation, with concentration in place of temperature, gives the diffusion equation for a spreading chemical, which is why both names sit on the very same mathematics.
Law one: heat flows downhill in temperature
The first ingredient is Fourier's law of heat conduction. It says the rate at which heat flows past a point is proportional to how steeply the temperature drops there, and it flows toward the cold side. Write the heat flux — energy crossing per unit time, taken positive when moving to the right — as q = -K u_x. The slope u_x measures how the temperature climbs as you move right; the minus sign turns a climb into a leftward push, so heat always heads downhill in temperature. The constant K > 0 is the conductivity, large for copper and small for wood. Notice the shape of this idea: it is the same downhill-gradient logic as Fick's law, where particles drift from crowded to sparse regions.
Law two: heat is only moved, never lost
The second ingredient is bookkeeping — a conservation law. Fence off a tiny slab of the bar from x to x + dx and tally its heat budget. Energy can change inside the slab only by crossing its two ends: it pours in at the left face and leaks out at the right. The stored heat is proportional to temperature, so the rate at which heat builds up inside is proportional to u_t times the slab's width dx. By conservation, that buildup must equal (flux in at left) minus (flux out at right) — anything that does not leave must still be here. Across the whole bar this is the statement that total heat is only redistributed, the conservation of total heat.
Now make the bookkeeping precise. The net inflow is q(x) - q(x + dx), the difference of flux at the two faces. For a small slab that difference is -q_x times dx, since q_x is just how fast the flux changes across space. Balancing buildup against net inflow gives (constant) u_t dx = -q_x dx. Cancel the common dx and the slab shrinks away, leaving a clean local law tying the time change of temperature to the spatial change of flux: c u_t = -q_x, where c bundles the material's density and heat capacity. This is the same conservation pattern u_t + q_x = (source) that you met for the transport equation — only the flux rule q is different.
Putting the two laws together
Two laws, one substitution. Conservation gave c u_t = -q_x; Fourier's law gave q = -K u_x. Differentiate the flux rule in x to get q_x = -K u_xx, then drop it into the conservation law: c u_t = -(-K u_xx) = K u_xx. Divide by c and define k = K / c. Out comes the heat equation, u_t = k u_xx. Every piece earned its place: the time derivative is the heat piling up, the second space derivative is the imbalance of flux between the two faces, and the constant k is what converts one into the other.
Fourier's law: q = -K u_x (flux runs downhill)
Conservation: c u_t = -q_x (nothing lost)
substitute q_x = -K u_xx:
c u_t = K u_xx
divide by c, set k = K/c:
u_t = k u_xx THE HEAT EQUATIONThat single positive number k = K / c is the thermal diffusivity, and it is worth a moment. It is not how hot the bar is or how much heat it holds — it is how quickly a temperature imbalance evens out. Large k (a metal) and bumps level off fast; small k (a brick wall) and they linger for hours. Reading the equation in words: the temperature at a point rises exactly when that point is cooler than the average of its immediate neighbours, because u_xx > 0 means the graph is cupped upward, a local dip. Heat floods into dips and drains out of bumps. That one sentence is the entire qualitative meaning of u_t = k u_xx.
What the equation already tells us — and what it doesn't
Look at where this equation sits in your classification. With one time derivative and two space derivatives it is parabolic, the borderline case between elliptic and hyperbolic — and that type is destiny. The single time derivative means time runs one way: the equation is not symmetric under t going to -t, which is the mathematical root of the arrow of time you will study two guides from now. The two space derivatives, second order and smoothing, foretell that sharp initial bumps will be instantly rounded off. The wave equation u_tt = c^2 u_xx, by contrast, has two time derivatives, is hyperbolic, and behaves utterly differently — finite speed, no smoothing. Same-looking symbols, opposite physics; the type is what decides.
Two honest caveats before we leave the derivation. First, we treated K, c, and k as constants — a clean assumption that holds for a uniform bar at modest temperature ranges. If the material varies in space, or k depends on temperature, the equation becomes variable-coefficient or nonlinear, and the tidy u_t = k u_xx is only the leading-order story. Second, this whole argument assumed conduction alone, with the bar's sides insulated; add a source of heat along the bar and conservation picks up an extra term, u_t = k u_xx + (source), the inhomogeneous case that the last guide of this rung handles with Duhamel's principle. The derivation is exact for what it assumes — and naming those assumptions is part of understanding it.