Understanding Integrating Factors¶

What is an Integrating Factor?¶

An integrating factor is a function that, when multiplied to a differential equation, makes it easier to solve. It's a powerful technique for solving linear differential equations (both ODEs and SDEs).

The Core Idea¶

Given a differential equation that's hard to solve directly, we multiply both sides by a cleverly chosen function \(\mu(t)\) that transforms it into a form we can integrate easily.

Motivating Example: First-Order Linear ODE¶

Consider the ODE:

\[ \frac{dy}{dt} + a(t)y = b(t) \]

This is hard to solve because \(y\) and \(\frac{dy}{dt}\) are mixed together.

The Trick: Multiply by an Integrating Factor¶

Define:

\[ \mu(t) = \exp\left(\int a(t)\,dt\right) \]

Key property: This \(\mu(t)\) satisfies:

\[ \frac{d\mu}{dt} = a(t)\mu(t) \]

Why This Helps¶

Multiply the original ODE by \(\mu(t)\):

\[ \mu(t)\frac{dy}{dt} + a(t)\mu(t)y = \mu(t)b(t) \]

Now notice: The left side is the derivative of \(\mu(t)y(t)\)!

\[ \frac{d}{dt}(\mu y) = \mu\frac{dy}{dt} + y\frac{d\mu}{dt} = \mu\frac{dy}{dt} + a(t)\mu y \]

So the ODE becomes:

\[ \frac{d}{dt}(\mu y) = \mu(t)b(t) \]

This is now easy to solve! Just integrate both sides:

\[ \mu(t)y(t) = \int \mu(s)b(s)\,ds + C \]

Then solve for \(y(t)\).

Deriving the Key Property: \(\frac{d\mu}{dt} = a(t)\mu(t)\)¶

Step 1: Definition¶

\[ \mu(t) = \exp\left(\int_0^t a(s)\,ds\right) \]

Step 2: Apply the Fundamental Theorem of Calculus¶

The derivative of an integral with respect to its upper limit is:

\[ \frac{d}{dt}\int_0^t a(s)\,ds = a(t) \]

Step 3: Apply the Chain Rule¶

Since \(\mu(t) = \exp(u(t))\) where \(u(t) = \int_0^t a(s)\,ds\):

\[ \frac{d\mu}{dt} = \frac{d}{dt}\exp(u(t)) = \exp(u(t)) \cdot \frac{du}{dt} \]

\[ \frac{d\mu}{dt} = \mu(t) \cdot a(t) = a(t)\mu(t) \]

Result: \(\boxed{\frac{d\mu}{dt} = a(t)\mu(t)}\)

This is exactly what we need!

Why This Property is Useful¶

The property \(\frac{d\mu}{dt} = a(t)\mu(t)\) allows us to recognize that:

\[ \frac{d}{dt}(\mu y) = \mu\frac{dy}{dt} + y\frac{d\mu}{dt} = \mu\frac{dy}{dt} + a(t)\mu y \]

This matches the left side of our ODE after multiplying by \(\mu\):

\[ \mu\frac{dy}{dt} + a(t)\mu y \]

So we can rewrite the ODE as:

\[ \frac{d}{dt}(\mu y) = \mu b(t) \]

which is immediately integrable.

Application to Linear SDEs¶

For the linear SDE:

\[ dx = a(t)x\,dt + b(t)\,dw \]

we use a similar integrating factor:

\[ \mu(t) = \exp\left(-\int_0^t a(s)\,ds\right) \]

Note the negative sign! This is because we want to cancel the drift term.

Why the Negative Sign?¶

In the SDE case, we want \(\frac{d\mu}{dt} = -a(t)\mu(t)\) (not \(+a(t)\mu(t)\)).

Let's derive it:

\[ \mu(t) = \exp\left(-\int_0^t a(s)\,ds\right) \]

\[ \frac{d\mu}{dt} = \exp\left(-\int_0^t a(s)\,ds\right) \cdot \frac{d}{dt}\left(-\int_0^t a(s)\,ds\right) \]

\[ \frac{d\mu}{dt} = \mu(t) \cdot (-a(t)) = -a(t)\mu(t) \]

Result: \(\boxed{\frac{d\mu}{dt} = -a(t)\mu(t)}\)

Applying Itô's Lemma¶

For the product \(\mu(t)x(t)\), Itô's lemma gives:

\[ d(\mu x) = \mu\,dx + x\,d\mu + d\mu \cdot dx \]

Since \(d\mu = -a(t)\mu\,dt\) (deterministic), the cross term \(d\mu \cdot dx\) is zero (deterministic × stochastic has no quadratic variation).

Substitute \(dx = a(t)x\,dt + b(t)\,dw\):

\[ d(\mu x) = \mu(a(t)x\,dt + b(t)\,dw) + x(-a(t)\mu\,dt) \]

\[ d(\mu x) = \mu a(t)x\,dt + \mu b(t)\,dw - a(t)\mu x\,dt \]

The drift terms cancel:

\[ d(\mu x) = \mu(t)b(t)\,dw \]

This is the key step! The drift has been eliminated, leaving only the stochastic term.

Summary: The Integrating Factor Method¶

For ODEs¶

Given: \(\frac{dy}{dt} + a(t)y = b(t)\)

Define: \(\mu(t) = \exp\left(\int a(t)\,dt\right)\)
Property: \(\frac{d\mu}{dt} = a(t)\mu(t)\)
Multiply ODE by \(\mu\): \(\frac{d}{dt}(\mu y) = \mu b(t)\)
Integrate: \(\mu y = \int \mu b\,dt + C\)
Solve for \(y\): \(y = \frac{1}{\mu}\left(\int \mu b\,dt + C\right)\)

For SDEs¶

Given: \(dx = a(t)x\,dt + b(t)\,dw\)

Define: \(\mu(t) = \exp\left(-\int_0^t a(s)\,ds\right)\) (note the negative sign)
Property: \(\frac{d\mu}{dt} = -a(t)\mu(t)\)
Apply Itô's lemma: \(d(\mu x) = \mu b(t)\,dw\) (drift cancels)
Integrate: \(\mu(t)x(t) = x(0) + \int_0^t \mu(s)b(s)\,dw(s)\)
Solve for \(x\): \(x(t) = \frac{1}{\mu(t)}\left(x(0) + \int_0^t \mu(s)b(s)\,dw(s)\right)\)

Intuitive Understanding¶

Why Does It Work?¶

The integrating factor "absorbs" the problematic term. Think of it as:

Before: \(y\) and \(\frac{dy}{dt}\) are entangled
After: The derivative of a product (\(\mu y\)) equals something simple

It's like completing the square or substitution—a transformation that simplifies the problem.

Physical Analogy¶

Imagine you're tracking a particle's position while the coordinate system is also moving. The integrating factor is like switching to a coordinate system that moves with the drift, making the equation simpler.

Key Takeaways¶

Integrating factors transform hard equations into easy ones
For ODEs: \(\mu(t) = \exp(\int a(t)\,dt)\) gives \(\frac{d\mu}{dt} = a(t)\mu(t)\)
For SDEs: \(\mu(t) = \exp(-\int_0^t a(s)\,ds)\) gives \(\frac{d\mu}{dt} = -a(t)\mu(t)\)
The negative sign in SDEs is crucial—it ensures drift cancellation
The product rule (or Itô's lemma) makes the transformed equation integrable

References¶

Boyce & DiPrima: Elementary Differential Equations — Classic ODE textbook
Øksendal (2003): Stochastic Differential Equations — Chapter 5 on linear SDEs
Kloeden & Platen (1992): Numerical Solution of SDEs — Comprehensive reference