Convolution Theorem - Differential Equations

SECTION 1

Historical Context & Motivation

Imagine you are an engineer in the 1800s trying to predict how a bridge vibrates under the weight of a passing train. The forces change over time, the bridge responds to each push differently, and you need a way to combine those effects into a single answer. This is exactly the kind of problem that led mathematicians to develop convolution — a mathematical operation that blends two functions together to describe their combined effect over time.

The Laplace transform, developed in the late 18th century by Pierre-Simon Laplace, converts functions of time into functions of a complex variable s. This conversion simplifies differential equations into algebraic ones. However, a recurring challenge arose: how do you take the inverse Laplace transform of a product of two transformed functions? The Convolution Theorem provides the elegant answer.

1782 • Laplace Introduces His Transform

Pierre-Simon Laplace publishes work on integral transforms while studying probability and celestial mechanics. His transform converts time-domain functions into a new "frequency" domain, laying the groundwork for solving differential equations algebraically.

1807 • Fourier's Heat Equation

Joseph Fourier introduces his own integral transform to solve the heat equation. His work on decomposing functions into sines and cosines inspires further development of convolution ideas in signal analysis.

1820s • Convolution Integral Formalized

Mathematicians recognize that the integral ∫f(τ)g(t − τ)dτ captures how two processes interact over time. This "convolution integral" becomes a standard tool in mathematical analysis.

1910s–1930s • Convolution Theorem Proven & Applied

The formal theorem linking convolution in the time domain to multiplication in the transform domain is rigorously established. Engineers begin applying it to electrical circuits and control systems.

1960s–Present • Digital Signal Processing Era

With the rise of computers, the Convolution Theorem becomes essential in digital signal processing, image filtering, and audio engineering, powering technologies from smartphones to medical imaging.

The central question the Convolution Theorem answers is this: if you know the Laplace transforms of two functions f(t) and g(t), and you see their transforms multiplied together as F(s) × G(s), how do you get back to the time domain? The theorem tells us the answer is not simply f(t) × g(t) — instead, it involves a special integral called the convolution of the two functions.

SECTION 2

Core Principles & Definitions

Before diving into the theorem itself, you need to understand a few foundational ideas. These building blocks connect what you already know about Laplace transforms to the new concept of convolution.

Laplace Transform Recap

The Laplace transform converts a time-domain function f(t) into a frequency-domain function F(s) using the integral ℒ{f(t)} = ∫₀^∞

e^{−st}

f(t) dt. This turns calculus problems into algebra problems.

Multiplication ≠ Simple Product

When you multiply two Laplace transforms F(s) × G(s), the inverse transform is not f(t) × g(t). The relationship is more intricate — it's a convolution.

The Convolution Integral

The convolution of f(t) and g(t), written (f ∗ g)(t), equals ∫₀ᵗ f(τ)g(t − τ) dτ. This integral slides one function across the other, accumulating their overlap at each moment.

The Theorem Statement

If ℒ{f(t)} = F(s) and ℒ{g(t)} = G(s), then ℒ{(f ∗ g)(t)} = F(s) × G(s). Equivalently, ℒ⁻¹{F(s) × G(s)} = (f ∗ g)(t). Products in s-space become convolutions in t-space.

Commutativity of Convolution

Convolution is commutative: f ∗ g = g ∗ f. This means ∫₀ᵗ f(τ)g(t − τ) dτ = ∫₀ᵗ g(τ)f(t − τ) dτ. You can choose whichever order makes the integral easier to evaluate.

✦

KEY TAKEAWAY

Think of convolution like making a smoothie. You don't just stack the ingredients on top of each other (simple multiplication). Instead, you blend them together — each ingredient affects the final taste based on how much of it you add at each moment. The convolution integral "blends" two functions by sliding one across the other and measuring their combined effect at every point in time. The Convolution Theorem tells you that this blending in the time domain is the same as simple multiplication in the s-domain.

SECTION 3

Visual Explanation

The diagram below shows the central idea of the Convolution Theorem. On the left side, you see two time-domain functions f(t) and g(t). In the time domain, combining them requires the convolution integral — a sliding, blending operation. However, if you first apply the Laplace transform (the downward arrows), you move into the s-domain where the combination is just ordinary multiplication. Then you take the inverse transform to return to the time domain with your answer.

The diagram shows two equivalent paths. The blue arrows represent Laplace transforms going down into the s-domain, where the functions are simply multiplied. The amber arrow shows the inverse transform returning to the time domain, where the result is the convolution integral.

Notice the key insight: there are two routes from the original functions to the combined result. You could try to evaluate the convolution integral directly in the time domain (the dashed box at the bottom left), which is often difficult. Or you could take the scenic route — transform to the s-domain, multiply, then inverse-transform. Both routes give you the exact same answer. The theorem guarantees it.

SECTION 4

Mathematical Framework

Let's formalize the mathematics. We'll define the convolution integral precisely, then state the theorem, and finally look at the key properties that make convolution manageable.

CONVOLUTION INTEGRAL

(f ∗ g)(t) = ∫₀ᵗ f(τ) · g(t − τ) dτ

f(τ) = the first function evaluated at dummy variable τ; g(t − τ) = the second function shifted and reflected; τ (tau) = integration variable running from 0 to t; t = the time at which you're evaluating the result.

CONVOLUTION THEOREM

ℒ{(f ∗ g)(t)} = F(s) × G(s)

Equivalently: ℒ⁻¹{F(s) × G(s)} = (f ∗ g)(t). The Laplace transform of a convolution equals the product of the individual transforms. This works in reverse too — the inverse transform of a product gives you the convolution.

The variable τ (tau) is a dummy variable of integration. When you compute (f ∗ g)(t), you hold t fixed and let τ sweep from 0 to t. At each value of τ, you multiply f(τ) by g(t − τ), then add up all those products. The expression g(t − τ) represents a version of g that has been flipped (reflected about τ = 0) and then shifted to the right by t units.

COMMUTATIVITY

(f ∗ g)(t) = (g ∗ f)(t)

You may swap f and g in the convolution integral. Choose whichever order leads to an easier integral.

DISTRIBUTIVITY

f ∗ (g + h) = (f ∗ g) + (f ∗ h)

Convolution distributes over addition, just like ordinary multiplication. This property is useful when dealing with sums of simpler functions.

⚠

Common Mistake Alert

Students often confuse the convolution integral's limits. Because we're working with Laplace transforms, where functions are defined for t ≥ 0, the limits of integration go from 0 to t (not from −∞ to ∞, which you might see in a Fourier transform context). Also remember: ℒ⁻¹{F(s) × G(s)} ≠ f(t) × g(t). The inverse of a product is a convolution, not a simple product!

SECTION 5

How Convolution Works Step by Step

To truly understand convolution, it helps to visualize what happens inside the integral. The following diagram illustrates the "flip and slide" process for two simple functions. We'll use f(τ) = τ and g(τ) = 1 (a constant), and compute (f ∗ g)(t) at a specific time t.

Panel A shows the two original functions. Panel B shows g reflected (flipped) about τ = 0 and slid to position t — since g(τ) = 1 is constant, the flip doesn't change its shape. Panel C shows the shaded overlap region whose area equals the value of (f ∗ g)(t). Here, the area of the triangle is t²/2.

Let's verify this with the Convolution Theorem. We know that ℒ{t} = 1/s² and ℒ{1} = 1/s. According to the theorem, ℒ{(f ∗ g)(t)} = F(s) × G(s) = (1/s²)(1/s) = 1/s³. Looking up the inverse Laplace transform, ℒ⁻¹{1/s³} = t²/2, which matches what we found from the integral. Both paths give the same answer, just as the theorem promises.

f(t)	g(t)	F(s) × G(s)	(f ∗ g)(t)
1	1	(1/s)(1/s) = 1/s²	t
t	1	(1/s²)(1/s) = 1/s³	t²/2
1	$e^{at}$	(1/s)·1/(s−a)	( $e^{at}$ − 1)/a
$e^{at}$	$e^{bt}$	1/((s−a)(s−b))	( $e^{at}$ − $e^{bt}$ )/(a − b)
t	t	(1/s²)(1/s²) = 1/s⁴	t³/6

Common convolution pairs and their Laplace transform products

SECTION 6

Worked Example

Let's work through a complete example. Suppose you need to find the inverse Laplace transform of H(s) = 1/[s²(s − 3)]. This expression doesn't match any single entry in a standard Laplace transform table, but the Convolution Theorem lets you split it into a product of two transforms you do recognize.

Finding ℒ⁻¹{1/[s²(s − 3)]} Using the Convolution Theorem

1
Step 1 — Split the Product

Rewrite H(s) as a product of two recognizable transforms: H(s) = F(s) × G(s) where F(s) = 1/s² and G(s) = 1/(s − 3). From a standard Laplace table, ℒ⁻¹{1/s²} = t and ℒ⁻¹{1/(s − 3)} = $e^{3t}$ .

f(t) = t, g(t) = $e^{3t}$
2
Step 2 — Set Up the Convolution Integral

By the Convolution Theorem, ℒ⁻¹{F(s) × G(s)} = (f ∗ g)(t) = ∫₀ᵗ f(τ) · g(t − τ) dτ. Substituting our functions: ∫₀ᵗ τ · $e^{3(t − τ)}$ dτ.

∫₀ᵗ τ · $e^{3(t − τ)}$ dτ
3
Step 3 — Simplify the Integrand

Factor out $e^{3t}$ since it doesn't depend on τ: $e^{3t}$ ∫₀ᵗ τ · $e^{−3τ}$ dτ. Now you have a standard integral of the form ∫ τ · $e^{−3τ}$ dτ, which can be evaluated using integration by parts.

$e^{3t}$ ∫₀ᵗ τ · $e^{−3τ}$ dτ
4
Step 4 — Integration by Parts

Let u = τ, dv = $e^{−3τ}$ dτ. Then du = dτ, v = −(1/3) $e^{−3τ}$ . By parts: ∫₀ᵗ τ $e^{−3τ}$ dτ = [−(τ/3) $e^{−3τ}$ ]₀ᵗ + (1/3)∫₀ᵗ $e^{−3τ}$ dτ = −(t/3) $e^{−3t}$ + (1/3)[−(1/3) $e^{−3τ}$ ]₀ᵗ = −(t/3) $e^{−3t}$ − (1/9) $e^{−3t}$ + 1/9.

∫₀ᵗ τ $e^{−3τ}$ dτ = −(t/3) $e^{−3t}$ − (1/9) $e^{−3t}$ + 1/9
5
Step 5 — Multiply by $e^{3t}$ and Simplify

Multiply the result from Step 4 by $e^{3t}$ : h(t) = $e^{3t}$ [−(t/3) $e^{−3t}$ − (1/9) $e^{−3t}$ + 1/9] = −t/3 − 1/9 + (1/9) $e^{3t}$ . Rearranging: h(t) = (1/9) $e^{3t}$ − (1/3)t − 1/9.

h(t) = (1/9) $e^{3t}$ − (1/3)t − 1/9
6
Step 6 — Verify (Optional)

You can check this answer using partial fractions. Decompose 1/[s²(s − 3)] = A/s + B/s² + C/(s − 3). Solving gives A = −1/9, B = −1/3, C = 1/9. Inverse-transforming each term: (1/9) $e^{3t}$ − (1/3)t − (1/9), which matches our convolution result exactly.

✓ Verified — both methods agree.

SECTION 7

Convolution vs. Partial Fractions

You might wonder: if partial fractions can also handle products in the s-domain, why bother with convolution? Each method has strengths and limitations. The table below compares them.

Feature	Convolution Theorem	Partial Fractions
Best when...	F(s) and G(s) have simple, recognizable inverse transforms	H(s) is a rational function with factorable denominator
Handles irrational H(s)?	Yes — works even when H(s) is not a ratio of polynomials	No — requires polynomial numerator and denominator
Difficulty	Requires evaluating an integral (possibly with integration by parts)	Requires solving a system of equations for coefficients
Answer form	Often gives a compact integral expression; may need simplification	Gives a sum of standard inverse transforms
Use in applications	Essential for systems theory, signal processing, and integral equations	Common in textbook ODE problems

Convolution Theorem vs. Partial Fractions for inverse Laplace transforms

✦

KEY TAKEAWAY

Think of partial fractions and convolution as two tools in your toolbox. Partial fractions are like a wrench — great for standard nuts and bolts (rational functions). Convolution is like a universal adapter — it works on a broader range of problems, including ones where partial fractions simply can't be applied. The Convolution Theorem is especially powerful in engineering and physics, where you often model a system's response to an input as a convolution of the input signal with the system's impulse response.

SECTION 8

Connections to Advanced Topics

The Convolution Theorem is not just a trick for textbook problems — it's a gateway concept that appears throughout higher mathematics, engineering, and science. Understanding it now gives you a head start on several advanced topics you may encounter later in your studies.

This Lesson	Advanced Topic
Convolution with Laplace transforms	Fourier convolution theorem — the same idea applies to Fourier transforms, which are central to signal processing and quantum mechanics
Convolution integral (continuous)	Discrete convolution — used in digital filters, neural networks (CNNs), and computer graphics
Solving ODEs with convolution	Green's functions — a generalized approach to solving PDEs using convolution with a fundamental solution
Product F(s) × G(s)	Transfer functions — in control theory, H(s) = Output(s)/Input(s) describes system behavior, and convolution models the output

How the Convolution Theorem connects to advanced concepts

One particularly exciting application is in convolutional neural networks (CNNs), which power modern image recognition and artificial intelligence. These networks apply discrete convolutions to filter and extract features from images — the same mathematical idea of "sliding and multiplying" that you've learned here, just adapted to two-dimensional grids of pixels. If you pursue engineering, computer science, or physics, the Convolution Theorem will reappear in many different forms.

ℹ

Looking Ahead

In a follow-up differential equations or signals course, you'll also learn about the Dirac delta function δ(t), which acts as an identity for convolution: f ∗ δ = f. This means convolving any function with a perfect "impulse" simply returns the original function — a beautiful and powerful property.

SECTION 9

Practice Problems

Test your understanding with these five problems, arranged from conceptual to challenging. Try each one before reading the answer.

Question 1: CONCEPTUAL

Explain in your own words why ℒ⁻¹{F(s) × G(s)} is not equal to f(t) × g(t). Why does convolution appear instead of simple multiplication?

Show answer

The Laplace transform involves an integral, not a simple algebraic operation. When you multiply two transforms F(s) and G(s), you're combining the effects of two integral operations, which interact in a more complex way than pointwise multiplication. The result in the time domain must account for how f and g influence each other across all previous moments — this accumulated interaction is what the convolution integral captures. Convolution "blends" the two functions by sliding one over the other and summing their products at each time step, which is fundamentally different from multiplying them at the same instant.

Question 2: BASIC CALCULATION

Use the Convolution Theorem to find ℒ⁻¹{1/[s(s − 2)]}. Identify f(t) and g(t) from the factors 1/s and 1/(s − 2), then evaluate the convolution integral.

Show answer

Write 1/[s(s − 2)] = F(s) × G(s) where F(s) = 1/s and G(s) = 1/(s − 2). From the Laplace table, f(t) = 1 and g(t) =

e^{2t}

. The convolution is (f ∗ g)(t) = ∫₀ᵗ 1 ·

e^{2(t−τ)}

dτ =

e^{2t}

∫₀ᵗ

e^{−2τ}

dτ =

e^{2t}

× [−(1/2)

e^{−2τ}

]₀ᵗ =

e^{2t}

× [−(1/2)

e^{−2t}

+ 1/2] = −1/2 + (1/2)

e^{2t}

= (

e^{2t}

− 1)/2. You can verify with partial fractions: 1/[s(s−2)] = −(1/2)/s + (1/2)/(s−2), giving ℒ⁻¹ = −1/2 + (1/2)

e^{2t}

, which matches.

Question 3: INTERMEDIATE

Find ℒ⁻¹{1/[s²(s + 1)]} using the Convolution Theorem. Hint: split into 1/s² and 1/(s + 1).

Show answer

Let F(s) = 1/s² (so f(t) = t) and G(s) = 1/(s + 1) (so g(t) =

e^{−t}

). Convolution: (f ∗ g)(t) = ∫₀ᵗ τ ·

e^{−(t−τ)}

dτ =

e^{−t}

∫₀ᵗ τ ·

e^{τ}

dτ. Integration by parts with u = τ, dv =

e^{τ}

dτ gives: ∫₀ᵗ τ

e^{τ}

dτ = [τe^τ]₀ᵗ − ∫₀ᵗ

e^{τ}

dτ =

te^t

−

e^t

+ 1. Multiplying by

e^{−t}

: h(t) =

e^{−t}

(

te^t

−

e^t

+ 1) = t − 1 +

e^{−t}

. So ℒ⁻¹{1/[s²(s + 1)]} = t − 1 +

e^{−t}

. This can be confirmed via partial fractions: 1/[s²(s+1)] = −1/s + 1/s² + 1/(s+1).

Question 4: APPLIED

A spring-mass system has transfer function H(s) = 1/(s² + 4), and the input force has Laplace transform X(s) = 1/s. The output Y(s) = H(s) × X(s) = 1/[s(s² + 4)]. Use convolution to find the time-domain output y(t).

Show answer

Identify f(t) = ℒ⁻¹{1/s} = 1 and g(t) = ℒ⁻¹{1/(s² + 4)} = (1/2)sin(2t). Convolution: y(t) = ∫₀ᵗ 1 · (1/2)sin(2(t − τ)) dτ = (1/2) ∫₀ᵗ sin(2t − 2τ) dτ. Let u = 2t − 2τ, so du = −2 dτ. When τ = 0, u = 2t; when τ = t, u = 0. The integral becomes (1/2)(−1/2) ∫₂ₜ⁰ sin(u) du = (1/4) ∫₀²ᵗ sin(u) du = (1/4)[−cos(u)]₀²ᵗ = (1/4)(1 − cos(2t)). So y(t) = (1 − cos(2t))/4. This represents the oscillating response of the spring-mass system to a constant force — the mass oscillates around a new equilibrium position.

Question 5: CRITICAL THINKING

The convolution of f(t) = sin(t) with itself is (f ∗ f)(t) = ∫₀ᵗ sin(τ)sin(t − τ) dτ. Without evaluating the integral directly, use the Convolution Theorem and the Laplace transform of sin(t) to find (sin ∗ sin)(t). Then reflect: does the result still look like a sine wave? Why or why not?

Show answer

The Laplace transform of sin(t) is F(s) = 1/(s² + 1). By the Convolution Theorem, ℒ{sin ∗ sin} = [1/(s² + 1)]² = 1/(s² + 1)². From the Laplace table (or by partial fractions), ℒ⁻¹{1/(s² + 1)²} = (1/2)(sin(t) − t·cos(t)). This result is not a simple sine wave. The term −(t/2)cos(t) has an amplitude that grows linearly with time, creating a phenomenon similar to resonance. Physically, this makes sense: convolving sin(t) with itself is like feeding a sine-wave input into a system that naturally oscillates at the same frequency, causing the response to grow over time. The convolution captures this cumulative reinforcement.

SUMMARY

Lesson Summary

Summary

The Convolution Theorem establishes that the Laplace transform of a convolution equals the product of the individual transforms: ℒ{(f ∗ g)(t)} = F(s) × G(s). The convolution integral ∫₀ᵗ f(τ)g(t − τ) dτ captures the accumulated interaction of two functions over time, using a "flip and slide" process where one function is reflected and shifted across the other.

This theorem provides a powerful alternative to partial fractions for finding inverse Laplace transforms of products. Convolution is commutative (f ∗ g = g ∗ f) and distributive over addition. Beyond differential equations, the Convolution Theorem is foundational in signal processing, control systems, and modern technologies like neural networks — making it one of the most far-reaching ideas you'll encounter in mathematics.