SDEs and Ito Intuition

How stochastic differential equations combine drift and Brownian noise, why Ito calculus replaces ordinary chain rules, and how Euler-Maruyama turns diffusion dynamics into stepwise simulation.

Keywords

stochastic differential equations, Ito calculus, drift, diffusion, Euler-Maruyama

1 Role

This is an early live page in the Stochastic Processes module.

Its job is to turn Brownian motion from a standalone random path into a reusable modeling language:

• deterministic local drift
• random local fluctuation
• continuous-time state evolution

This is the point where diffusion intuition becomes an equation.

2 First-Pass Promise

You can read this page on its own inside the full module spine.

If you stop here, you should still understand:

• what an SDE looks like at first pass
• how drift and diffusion play different roles
• why ordinary chain-rule intuition breaks
• why Ito formula adds a second-order term
• how Euler-Maruyama approximates an SDE by discrete steps

3 Why It Matters

Ordinary differential equations say:

state change = deterministic local rule

SDEs say:

state change = deterministic local rule + continuous random fluctuation

That extra noise term is the backbone behind:

• stochastic control
• continuous-time filtering
• diffusion and transport models
• random physical or financial dynamics
• PDE bridges like Fokker-Planck or Feynman-Kac viewpoints

So this page matters because it explains how Brownian noise enters an actual state equation.

4 Prerequisite Recall

• from Probability, Gaussian noise is characterized by mean and variance
• from Brownian Motion and Diffusion Intuition, Brownian increments over length dt behave like centered Gaussian noise with variance dt
• from ODE intuition nearby, a differential equation gives a local rule for evolution over small time intervals

5 Intuition

5.1 An SDE Is A Drift-Plus-Noise Local Rule

At first pass, an SDE has the form

\[ dX_t = b(X_t,t)\,dt + \sigma(X_t,t)\,dW_t. \]

Read it like this:

• b(X_t,t) dt is the deterministic local push
• \sigma(X_t,t) dW_t is the random local fluctuation

So the state changes for two different reasons:

• structure from the drift
• uncertainty from Brownian noise

5.2 Drift And Diffusion Live On Different Scales

Over a short interval of length dt, Brownian motion changes by size about \sqrt{dt}.

That means:

• drift contributes on scale dt
• noise contributes on scale \sqrt{dt}

This is why noise cannot be treated like an ordinary small deterministic perturbation.

5.3 Ordinary Chain Rules No Longer Fit

If dW_t were just an ordinary differential, we might expect standard chain-rule behavior.

But Brownian paths are too rough.

The first-pass slogan is:

• dW_t is size \sqrt{dt}
• so (dW_t)^2 behaves like dt

That is the source of the extra second-order term in Ito formula.

5.4 Simulation Needs A Stochastic Time-Step Rule

Even though the model is continuous-time, computation still uses steps.

The simplest first-pass simulation rule is:

\[ X_{k+1} \approx X_k + b(X_k,t_k)\Delta t + \sigma(X_k,t_k)\sqrt{\Delta t}\,Z_k, \]

where Z_k \sim N(0,1).

This is Euler-Maruyama.

It is the stochastic analogue of Euler time-stepping for ODEs.

6 Formal Core

Definition 1 (Definition: Stochastic Differential Equation) At a first pass, an SDE is a continuous-time state evolution rule of the form

\[ dX_t = b(X_t,t)\,dt + \sigma(X_t,t)\,dW_t, \]

where:

• b is the drift
• \sigma scales the diffusion noise
• W_t is Brownian motion

Theorem 1 (Theorem Idea: Short-Time SDE Increments) Over a short interval of length \Delta t,

\[ X_{t+\Delta t} - X_t \approx b(X_t,t)\Delta t + \sigma(X_t,t)\sqrt{\Delta t}\,Z, \]

with Z \sim N(0,1).

At first pass, this is the cleanest way to read the equation computationally and probabilistically at the same time.

Theorem 2 (Theorem Idea: Ito Formula) If X_t follows

\[ dX_t = b(X_t,t)\,dt + \sigma(X_t,t)\,dW_t \]

and f(t,x) is smooth enough, then

\[ df(t,X_t) = \partial_t f\,dt + f_x\,dX_t + \frac12 \sigma(X_t,t)^2 f_{xx}\,dt \]

in the scalar first-pass setting.

The important feature is the extra \frac12 \sigma^2 f_{xx} term.

That term is the visible signature that stochastic calculus is not ordinary calculus.

Theorem 3 (Theorem Idea: Euler-Maruyama Update) The simplest first-pass discretization of an SDE is

\[ X_{k+1} = X_k + b(X_k,t_k)\Delta t + \sigma(X_k,t_k)\sqrt{\Delta t}\,Z_k, \]

where Z_k are independent standard Gaussians.

This is the default bridge from continuous-time stochastic models to simulation.

7 Worked Example

Consider the constant-coefficient SDE

\[ dX_t = \mu\,dt + \sigma\,dW_t. \]

This says:

• there is deterministic drift with rate \mu
• plus Brownian fluctuation with scale \sigma

Now apply Ito intuition to the function f(x)=x^2.

Ordinary calculus might suggest:

\[ d(X_t^2) \approx 2X_t\,dX_t. \]

Ito formula says there is an extra correction:

\[ d(X_t^2) = 2X_t\,dX_t + \sigma^2\,dt. \]

Substituting dX_t gives

\[ d(X_t^2) = 2\mu X_t\,dt + 2\sigma X_t\,dW_t + \sigma^2 dt. \]

So the second moment does not evolve only because of drift.

It also receives a direct variance injection from the noise itself.

That extra \sigma^2 dt term is the first concrete reason Ito calculus matters.

8 Computation Lens

When you meet an SDE, ask:

1. what is the deterministic drift term?
2. what is the diffusion or noise term?
3. what scale does noise contribute over one time step?
4. does a nonlinear transformation require Ito formula instead of ordinary chain rules?
5. what numerical scheme is being used to simulate or approximate the process?

Those questions usually reveal whether the paper or model is emphasizing:

• stochastic dynamics
• uncertainty propagation
• continuous-time control
• or modern diffusion-style generation

9 Application Lens

9.1 Continuous-Time Stochastic Control

Controlled diffusions are written as SDEs with extra control inputs inside the drift and sometimes the diffusion term.

9.2 Filtering And State Estimation

Hidden-state models in continuous time often describe latent dynamics with SDEs and observations with noisy measurement rules.

9.3 Diffusion And Score-Based ML

Modern diffusion models reuse the same drift-plus-noise language, then study reverse-time SDEs or probability-flow ODEs for generation.

10 Stop Here For First Pass

If you stop here, retain these five ideas:

• an SDE is a local rule with deterministic drift plus Brownian noise
• drift scales like dt, while noise scales like \sqrt{dt}
• Brownian roughness is why ordinary chain rules fail
• Ito formula adds an extra second-order correction term
• Euler-Maruyama is the simplest simulation bridge from SDEs to discrete updates

11 Go Deeper

The strongest adjacent live pages right now are:

• Brownian Motion and Diffusion Intuition
• Mixing, Ergodicity, and MCMC Bridges
• Continuous-Time Stochastic Control and Hamilton-Jacobi-Bellman Intuition
• Score Matching and the SDE View of Diffusion
• Research Bridges: Reverse-Time SDEs, Probability-Flow ODEs, Flow Matching, and Control

The natural adjacent long-run page after this one is:

• Mixing, Ergodicity, and MCMC Bridges

12 Optional Deeper Reading After First Pass

The strongest current references connected to this page are:

• MIT 18.445 lecture notes page - official MIT route where Brownian motion, martingales, and continuous-time process ideas sit in one sequence. Checked 2026-04-25.
• MIT 15.070J lecture notes page - official MIT lecture hub that explicitly includes Ito calculus, Ito integrals, and Ito formula. Checked 2026-04-25.
• MIT 15.070J lecture 17 - official MIT lecture specifically centered on Ito processes and Ito formula. Checked 2026-04-25.
• Stanford Stats 310C syllabus - official Stanford syllabus showing the local route from Brownian motion to stochastic integrals, Ito formula, and Feynman-Kac ideas. Checked 2026-04-25.

13 Sources and Further Reading

• MIT 18.445 lecture notes page - First pass - official MIT lecture hub for stochastic-process foundations feeding into diffusion and stochastic-calculus viewpoints. Checked 2026-04-25.
• MIT 15.070J lecture notes page - First pass - official MIT lecture hub explicitly covering Ito calculus and Ito formula. Checked 2026-04-25.
• MIT 15.070J lecture 17 - Second pass - useful official MIT note once you want the Ito-process and Ito-formula object stated more directly. Checked 2026-04-25.
• Stanford Stats 218 - Second pass - official Stanford course page placing diffusion processes inside a broader stochastic-process curriculum. Checked 2026-04-25.
• Stanford Stats 310C - Second pass - official Stanford probability page where continuous-time stochastic processes are part of the course backbone. Checked 2026-04-25.
• Stanford Stats 310C syllabus - Bridge outward - useful Stanford syllabus showing the progression from Brownian motion to stochastic integrals and Ito formula. Checked 2026-04-25.