PDE Classification and Boundary Conditions

The Black–Scholes equation tells you how an option's price changes as both time and the underlying's price move — it's a rule about two variables at once, not one. Equations like this, involving rates of change with respect to more than one variable simultaneously, are partial differential equations (PDEs), and unlike ordinary differential equations they come in genuinely different families that behave nothing alike: some smooth out sharp features over time, some carry a shape along unchanged, some just balance forces at rest. Knowing which family you're in tells you what extra information — boundary conditions — the equation needs before it has one unique answer instead of infinitely many.

The analogy: heat, waves, and a stretched drumhead

Drop a hot poker into a bucket of water: the sharp temperature spike smooths out and spreads over time — that's diffusion (or "parabolic") behavior, and it's what the Black–Scholes equation does to an option payoff's kink as time runs forward. Pluck a guitar string: the disturbance travels along the string as a wave, keeping its shape rather than smoothing away — that's hyperbolic behavior. Now stretch a drumhead over a frame and let it settle: no time is involved at all, just a static balance of tension everywhere — that's elliptic behavior, describing an equilibrium rather than an evolution.

The mechanics, one symbol at a time

A general second-order linear PDE in two variables can be written $A\,u_{xx} + B\,u_{xy} + C\,u_{yy} + (\text{lower-order terms}) = 0$, where subscripts denote partial derivatives. Its type is determined purely by the discriminant:

$$\Delta = B^2 - 4AC.$$

In plain English: this single number, built from the coefficients on the second-derivative terms, sorts every such equation into exactly one of three families —

$$\Delta < 0 \Rightarrow \text{elliptic (equilibrium)}, \quad \Delta = 0 \Rightarrow \text{parabolic (diffusion)}, \quad \Delta > 0 \Rightarrow \text{hyperbolic (wave)}.$$

The Black–Scholes PDE, after the usual variable changes, is exactly the heat equation — parabolic. This matters for boundary conditions: a parabolic equation needs a condition at one end of time (the option's known payoff at expiry) plus conditions at the spatial edges (what the price does as spot goes to zero or infinity), and it can then be solved uniquely running backward in time from expiry to today.

Worked example 1: classifying Black–Scholes

The Black–Scholes PDE for an option price $V(S,t)$ is $\frac{\partial V}{\partial t} + \frac12\sigma^2 S^2\frac{\partial^2 V}{\partial S^2} + rS\frac{\partial V}{\partial S} - rV = 0$. Writing it in the $A u_{xx}+Bu_{xy}+Cu_{yy}$ form with $x=S,\,y=t$: $A = \tfrac12\sigma^2S^2$, $B=0$ (no mixed $S,t$ second derivative), $C=0$ (no $\partial^2/\partial t^2$ term). The discriminant is $\Delta = 0^2 - 4(\tfrac12\sigma^2S^2)(0) = 0$: parabolic, confirming it behaves like heat diffusion — smoothing the kinked payoff at expiry into a smooth curve as you move backward away from expiry.

Worked example 2: what boundary conditions a European call needs

To pin down a unique solution for a European call with strike $K=100$, maturity $T=1$, you need: (1) a terminal condition at $t=T$ — the known payoff $V(S,T) = \max(S-100, 0)$; (2) a boundary condition as $S \to 0$ — the call is worthless if the stock is worthless, so $V(0,t)=0$; (3) a boundary condition as $S \to \infty$ — deep in the money the call behaves like the forward, $V(S,t) \to S - Ke^{-r(T-t)}$. Concretely, at $t=T$ with $S=120$: V = \max(120-100,0) = \20$ exactly, by the terminal condition alone — no PDE solving needed at that single boundary point, since it's given, not derived. Miss any one of these three pieces and the same PDE has infinitely many equally valid solutions; the boundary conditions are what select the one that's actually the option price.

Think of the payoff curve at expiry as the starting shape (like $\max(S-K,0)$ near the kink) — this is the terminal condition a parabolic PDE like Black–Scholes smooths out as you solve backward in time.

What this means in practice

Recognizing a PDE's type before touching a numerical scheme is not academic: parabolic equations are typically solved with finite-difference schemes that step backward in time from a terminal condition (exactly how option pricing engines discretize Black–Scholes), while genuinely hyperbolic or elliptic problems need different solvers and different boundary setups entirely. Getting the boundary conditions wrong — even with the right PDE — is one of the most common sources of silent mispricing in a homegrown pricing engine, especially near the edges of the spot grid.