· Hankyu Kim · Estimation  · 3 min read

Ito Calculus and Ito's Lemma

Understanding Brownian motion, Wiener processes, and why Ito calculus is necessary for stochastic systems.

Understanding Brownian motion, Wiener processes, and why Ito calculus is necessary for stochastic systems.

Brownian Motion and Wiener Process

Brownian Motion describes the random movement of particles caused by collisions with surrounding molecules.

A Wiener Process W(t)W(t) is a mathematical model of Brownian motion.

A standard Wiener Process satisfies

W(0)=0W(0)=0 W(t)∼N(0,t)\boxed{W(t)\sim\mathcal{N}(0,t)}

Therefore,

E[W(t)]=0,Var⁡(W(t))=tE[W(t)]=0,\qquad \operatorname{Var}(W(t))=t

The expected value remains zero, while the variance increases with time.

For a time interval Δt\Delta t, the Wiener increment follows

ΔW∼N(0,Δt)\boxed{\Delta W\sim\mathcal{N}(0,\Delta t)}

Wiener increments over non-overlapping time intervals are independent.

Wiener Process

In MATLAB, a Wiener increment can be generated using

dw = sqrt(dt) * randn;

Starting with a Simple Position and Velocity Model

Suppose a vehicle moves with constant velocity.

dxdt=v\frac{dx}{dt}=v

Integrating,

x(t)=x(0)+vtx(t)=x(0)+vt

However, actual velocity may contain random fluctuations.

dxdt=v+ξ(t)\frac{dx}{dt}=v+\xi(t)

where ξ(t)\xi(t) represents Gaussian white noise.

Ideal Gaussian white noise cannot be integrated using conventional calculus because it is not an ordinary function.

Instead, we model its accumulated effect using Wiener increments.

Formally,

dW=ξ(t) dtdW=\xi(t)\,dt

The system becomes

dx=v dt+σ dW\boxed{dx=v\,dt+\sigma\,dW}

where:

  • v dtv\,dt: deterministic change
  • σ dW\sigma\,dW: stochastic change

This is a Stochastic Differential Equation (SDE).


Why Is (dW)2=dt(dW)^2=dt Important?

In conventional calculus, higher-order infinitesimal terms such as (dx)2(dx)^2 are neglected.

However, Wiener increments behave differently.

Since

ΔW∼N(0,Δt)\Delta W\sim\mathcal{N}(0,\Delta t)

we have

E[(ΔW)2]=ΔtE[(\Delta W)^2]=\Delta t

This leads to the fundamental Itô calculus rule:

(dW)2=dt\boxed{(dW)^2=dt}

This is a symbolic rule based on the quadratic variation of Wiener processes, not an exact equality for individual increments.

The Itô multiplication rules are

(dt)2=0,dt dW=0,(dW)2=dt(dt)^2=0,\qquad dt\,dW=0,\qquad(dW)^2=dt

Unlike conventional calculus, the second-order Wiener term cannot be ignored.

This produces the additional term in Itô’s Lemma.


What Is Ito Integration?

Consider a stochastic differential equation:

dx=f(x,t) dt+b(x,t) dW\boxed{dx=f(x,t)\,dt+b(x,t)\,dW}

where f(x,t)f(x,t) describes deterministic dynamics and b(x,t)b(x,t) determines the noise intensity.

For example,

dx=−kx dt+σx dWdx=-kx\,dt+\sigma x\,dW

Integrating,

x(T)=x(0)−∫0Tkx(t) dt+σ∫0Tx(t) dW(t)x(T)=x(0)-\int_0^T kx(t)\,dt+\sigma\int_0^T x(t)\,dW(t)

The last term is a stochastic integral.

Itô integration defines how to integrate with respect to Wiener increments.

∫0Tx(t) dW(t)=lim⁡Δt→0∑kx(tk)ΔWk\boxed{ \int_0^T x(t)\,dW(t) = \lim_{\Delta t\to0}\sum_k x(t_k)\Delta W_k }

The integrand is evaluated at the beginning of each time interval, and the limit is understood in a stochastic sense.


What Is Ito’s Lemma?

Itô’s Lemma is the Chain Rule for stochastic processes.

Suppose

dx=dWdx=dW

and define

y=x2y=x^2

In conventional calculus,

dy=2x dxdy=2x\,dx

However, consider the exact change:

Δy=(x+Δx)2−x2\Delta y=(x+\Delta x)^2-x^2

Expanding,

Δy=2xΔx+(Δx)2\Delta y=2x\Delta x+(\Delta x)^2

Since Δx=ΔW\Delta x=\Delta W,

Δy=2xΔW+(ΔW)2\Delta y=2x\Delta W+(\Delta W)^2

Using the Itô rule (dW)2=dt(dW)^2=dt,

dy=2x dW+dt\boxed{dy=2x\,dW+dt}

The additional dtdt is called the Itô Correction Term.


General Form of Ito’s Lemma

Suppose

dx=f(x,t) dt+b(x,t) dWdx=f(x,t)\,dt+b(x,t)\,dW

and

y=g(x,t)y=g(x,t)

Then Itô’s Lemma gives

dy=(∂g∂t+f∂g∂x+12b2∂2g∂x2)dt+b∂g∂xdW\boxed{ dy= \left( \frac{\partial g}{\partial t} +f\frac{\partial g}{\partial x} +\frac12b^2\frac{\partial^2g}{\partial x^2} \right)dt + b\frac{\partial g}{\partial x}dW }

Compared with the ordinary Chain Rule, the additional term is

12b2∂2g∂x2dt\boxed{ \frac12b^2\frac{\partial^2g}{\partial x^2}dt }

This is the Itô Correction Term, which arises because (dW)2=dt(dW)^2=dt.

Itô’s Lemma allows us to calculate how a function of a stochastic state evolves.


Summary

  • Brownian Motion: Random particle movement.
  • Wiener Process: Mathematical model of Brownian motion.
  • Stochastic Differential Equation: Dynamic system with random noise.
  • Itô Integration: Integration with respect to Wiener increments.
  • Itô’s Lemma: Chain Rule for stochastic processes.
  • Itô Correction Term: Additional term caused by the quadratic variation of Wiener processes.

The key relationship is

(dW)2=dt\boxed{(dW)^2=dt}

Wiener processes model random disturbances, Itô integration defines their accumulation, and Itô’s Lemma describes how functions of stochastic states evolve.


Back to Blog

Related Posts

View All Posts »
Extended Kalman Filter (EKF)

Extended Kalman Filter (EKF)

The Extended Kalman Filter applies the Kalman filter to nonlinear systems by locally linearizing system and measurement models using Jacobians.

Kalman Filter (Part 1)

Kalman Filter (Part 1)

The Kalman filter is an optimal recursive estimator for linear systems, combining system models and noisy measurements through simple matrix operations.

Kalman Filter (Part 2)

Kalman Filter (Part 2)

This post explains the physical meaning of the error covariance P and the Kalman gain K, showing how the Kalman filter adaptively balances model prediction and sensor measurements.

Low Pass Filter (LPF)

Low Pass Filter (LPF)

A low pass filter reduces noise while emphasizing recent measurements, overcoming the limitations of uniform averaging in moving average filters.