JOVANA
Library Glossary Getting Started Three Levels Fields How it works Mission
Join the mission
All guides

The Three Model Equations

Heat spreading, waves travelling, a membrane settling into stillness — these are the three faces of second-order PDEs. Meet the heat, wave, and Laplace/Poisson equations, see what each one models, and learn the one rule that decides what data each one is allowed to ask of you.

Three equations, one Laplacian, three personalities

In the previous guide you learned to sort a second-order partial differential equation into one of three boxes — parabolic, hyperbolic, or elliptic — by reading the sign of a discriminant, exactly the way the discriminant of a conic tells a parabola from a hyperbola from an ellipse. That classification was not bookkeeping for its own sake. Each box has one clean representative, a model equation that every other equation in the box behaves like, and those three representatives are the subject of this whole rung. Meet them once, deeply, and the rest of PDE theory becomes variations on three themes.

Here is a startling unity to hold onto first: all three are built from the very same spatial operator, the Laplacian nabla^2 u, which in one dimension is just d^2u/dx^2 and in general is the sum of the unmixed second partial derivatives, d^2u/dx^2 + d^2u/dy^2 + d^2u/dz^2. The Laplacian measures how much the value at a point falls below the average of its immediate neighbours — it is curvature, generalized to many directions. What separates the three equations is not the space part at all. It is entirely a question of what sits on the LEFT, in time: one time derivative, two time derivatives, or none.

THE THREE MODEL EQUATIONS  (u = u(x, t)  or  u = u(x, y))

  HEAT / diffusion   (parabolic)   du/dt   = k * nabla^2 u
  WAVE               (hyperbolic)   d^2u/dt^2 = c^2 * nabla^2 u
  LAPLACE            (elliptic)     nabla^2 u = 0
  POISSON            (elliptic)     nabla^2 u = f      (= source)

  Same Laplacian on the right.  The TIME side decides everything:
     one  d/dt   -> smoothing, irreversible, settles forever  (heat)
     two  d^2/dt^2 -> oscillation, reversible, travels at speed c (wave)
     none           -> the steady, settled state itself        (Laplace)
The whole rung in one box. Notice that Laplace's equation is literally the heat or wave equation with the time derivative switched off — it is what either one relaxes to once nothing changes in time.

Heat: smoothing, forgetting, one-way time

The heat equation du/dt = k nabla^2 u says, in plain words: a point's temperature rises in time exactly in proportion to how much colder it is than its neighbours' average. That is the Laplacian on the right, read as a thermometer of local imbalance. Where you are colder than your surroundings (positive Laplacian), you warm up; where you are a hot spike (negative Laplacian), you cool down. The net effect is relentless averaging: bumps flatten, dips fill in, and any sharp feature smears out. Drop ink in still water, or hold one end of a metal rod to a flame — the same equation, with k the diffusivity, governs how the disturbance spreads.

Because there is only ONE time derivative, the equation needs exactly one snapshot in time to get going: the initial condition u(x, 0), the temperature everywhere at the starting instant. This is the same logic as a first-order ODE from Volume I, where dy/dt = f only needed a single starting value y(0) — one derivative in time, one initial datum. You do not, and must not, also specify du/dt at t = 0; the equation itself computes that for you from the spatial shape. Add to this the boundary conditions in space (what the rod's ends are doing for all time), and the future is fully determined.

Wave: oscillation, memory, two-way time

The wave equation d^2u/dt^2 = c^2 nabla^2 u looks almost identical, but the single change — a SECOND time derivative on the left — flips the personality completely. Read it as Newton's law for a stretched string: d^2u/dt^2 is the acceleration of the bit of string at position x, and the Laplacian on the right is the restoring force, because a bit of string that sits below its neighbours' average gets pulled up by the tension. Acceleration toward the average, not velocity toward it, is the recipe for oscillation rather than decay — the same difference as between a damped slide to rest and a mass on a spring that overshoots and swings back. The displacement does not flatten out; it sloshes, with c the wave speed.

Two time derivatives mean two initial conditions, and now the analogy is a second-order ODE like a mass on a spring, which needed both an initial position AND an initial velocity. For the string you must supply u(x, 0), the initial shape (how you plucked it), AND du/dt(x, 0), the initial velocity (whether you also struck it moving). Give only the shape and the future is genuinely undetermined — a guitar string plucked and released behaves differently from the same shape launched with a sideways flick. Both data are physically real and both are mathematically required.

Two deeper contrasts with heat are worth naming. First, the wave equation is reversible: it is unchanged if you replace t by -t (the second derivative does not notice the sign flip), so a film of a vibrating string looks just as physical run backward — no arrow of time, and no smearing-out of detail. Second, signals travel at a finite speed c. A disturbance at one spot is not felt elsewhere until a wave physically arrives; on the infinite line this is captured exactly by d'Alembert's solution, which splits any initial shape into two copies gliding off left and right at speed c. Heat, by contrast, technically responds everywhere instantaneously (a flaw of the idealized model, since real diffusion has a speed limit too).

Laplace and Poisson: the equation with no clock

Now switch the clock off entirely. Wait long enough and a heated plate stops changing — every point has reached the average of its neighbours, so the Laplacian is zero everywhere and du/dt has died. What remains is Laplace's equation nabla^2 u = 0, the equation of the steady state: the final, settled temperature of a plate whose edges are held fixed, the shape of a soap film stretched on a wire loop, the electrostatic potential in a charge-free region. Its solutions are called harmonic functions, and they have a beautiful defining property — the mean-value property: the value at any point equals the average of the values on any circle (or sphere) around it. A harmonic function is, quite literally, a function with no bumps of its own; it is as flat as its boundary will permit.

When the steady state is driven by a source that never switches off — a permanently heated wire inside the plate, an electric charge sitting in the region — the right side is no longer zero, and you get Poisson's equation nabla^2 u = f, where f records the source strength at each point. Laplace is just Poisson with the source turned off (f = 0). In electrostatics this is the master equation: nabla^2 (potential) = -(charge density)/(permittivity), the statement that charges are the dimples and spikes pinning down an otherwise flat potential landscape.

With no time variable there is no 'initial' anything — you cannot give a starting instant for an equation that describes a thing that already finished changing. Instead these equations live on a closed region and ask only for data on its entire boundary: this is a pure boundary-value problem. Pin the temperature all the way around the rim of the plate and the interior is then completely and uniquely determined. The maximum principle makes this vivid: a harmonic function attains its hottest and coldest values only on the boundary, never strictly inside — the steady plate has no interior hot spot the edges did not put there. That is why fixing the boundary alone suffices to fix everything within.

What data each one is allowed to ask for

Pull the pattern together, because it is the single most useful rule in this rung: the number of TIME derivatives tells you how many INITIAL conditions to supply, and the spatial boundary always needs conditions all the way around its edge for all time. Heat (one d/dt) wants one initial profile; wave (two d^2/dt^2) wants two — shape and velocity; Laplace and Poisson (no time) want no initial data at all, only the full boundary. Match the count exactly. Too few conditions and the answer is not pinned down; too many and you have over-constrained a system that physics says is already determined, and typically no solution exists at all.

The boundary conditions themselves come in three flavours, and the physics picks which. A Dirichlet condition fixes the VALUE on the edge — the rod's ends held at 0 degrees, the membrane clamped to a frame. A Neumann condition fixes the FLUX, the normal derivative du/dn — an insulated end through which no heat leaks (du/dn = 0), or a free, untethered edge. A Robin condition mixes the two, as when a surface radiates heat to the air at a rate proportional to its temperature. Which flavour you use is dictated by what the world is doing at that boundary, not by mathematical taste.

  1. Classify the equation: count its time derivatives (zero, one, or two) — this is the parabolic / hyperbolic / elliptic call from the previous guide.
  2. Supply that many initial conditions at t = 0: none for Laplace/Poisson, one (u) for heat, two (u and du/dt) for wave.
  3. Specify a boundary condition all the way around the spatial region, for all time — Dirichlet (value), Neumann (flux), or Robin (a mix), chosen by the physics of each edge.
  4. Check it is well-posed: a solution should exist, be unique, and depend continuously on the data — exactly the right amount of information, no more and no less.

Well-posedness, and the road from here

The phrase that ties all of this together is well-posedness, due to Hadamard, and it has three demanding parts. A problem is a well-posed problem if (1) a solution EXISTS, (2) it is UNIQUE, and (3) it depends CONTINUOUSLY on the data, so a small wobble in the initial or boundary values produces only a small wobble in the answer. The right boundary-and-initial data — matched to the equation's type by the counting rule above — is precisely what buys you all three. The wrong data breaks one of them, and the breakage is not a mere inconvenience; it usually means you have mis-modelled the physics.

You now have the cast of characters and the rule that governs their casting. What you do NOT yet have is a way to actually solve any of them — and that is the subject of the next guides, where separation of variables enters. The plan there is elegant: guess that the solution factors into a function of space times a function of time, watch the single PDE split into two ordinary differential equations you already know how to handle from Volume I, and then rebuild the full answer by superposing solutions into a series — a Fourier series, in fact, which is exactly why the previous rung on Fourier series came first. Be warned that separation of variables is not a universal solver; it works only on special, symmetric geometries with cooperative boundary conditions. But for our three model equations on a rod, a rectangle, or a disc, it is the master key, and you are now ready to turn it.