Stochastic differential equations · Part 6 of 10
From SDEs to PDEs
Expectations of functions of a diffusion solve partial differential equations, and solutions of many PDEs can be written as averages over diffusion paths. This post derives the dictionary between the two.
The generator
For the diffusion in one dimension, the generator is the differential operator
In dimensions, with , it is . For Brownian motion, .
The basic computation
Itô’s formula applied to a function reads
The last term is a (local) martingale. So if solves a PDE that makes the term vanish, then is a martingale, and taking expectations at two different times links the PDE to the process. Every result below is a variation on this one computation.
Terminal value problems: the backward equation
Fix a time and a function , and let
Then solves the backward Kolmogorov equation
Conversely, if is a sufficiently regular solution of this problem, the basic computation shows that is a martingale on , so . The equation runs backward in time because its condition is imposed at the final time.
Two extensions follow from the same argument:
- A source term. The solution of with is
- A potential (the Feynman–Kac formula). The solution of with is One interpretation is that the process is killed at rate , and the exponential is the probability that it survives until time . In finance, is an interest rate and the formula is a discounted expected payoff.
Initial value problems and long horizons
When the coefficients do not depend on time, the function , the expectation for the process started at , solves the forward-in-time problem
For Brownian motion this says that solves the heat equation , which is the probabilistic reading of convolution with the Gaussian heat kernel.
Over an infinite horizon, with discounting at a rate ,
Stopping times and boundary value problems
Let be a domain and the first time leaves it. Running the basic computation up to the stopping time , and applying the optional stopping theorem from the post on martingales, gives representations of elliptic boundary value problems:
- solves in , with on the boundary (the Dirichlet problem);
- solves in , with on .
For Brownian motion in a domain of , the first statement says that the harmonic function with boundary values , evaluated at , is the average of over the point where a Brownian motion started at first hits the boundary.
Example: Brownian motion on an interval
Start a Brownian motion at a point and stop it when it first hits or .
Which end does it reach first? The probability of reaching first solves with and , so
How long does it take? The expected exit time solves with , so
Starting in the middle, it takes on average. That is the diffusive scaling again: an interval twice as long takes four times as long to escape.
With drift. For , the generator is , and the same boundary value problem gives
As , this recovers . The function that appears here is the scale function of the process, which plays the starring role in the next post.
The forward equation
The backward equation describes expectations as functions of the starting point. The density of evolves instead by the adjoint operator, in the Fokker–Planck (or forward Kolmogorov) equation:
A stationary density solves . For the Langevin diffusion , the probability flux is identically zero, which is a one-line proof that is its stationary distribution.
Why the dictionary is useful
Read from right to left, these formulas turn a PDE into a Monte Carlo problem: simulate paths and average them, which works in dimensions where grid-based solvers are hopeless. Read from left to right, they turn probabilistic questions about exit times, hitting probabilities and prices into differential equations that can sometimes be solved exactly.