What is an intuitive picture of convolution?

Think of sliding one function across another and, at each position, multiplying them point by point and adding up the result. The output measures how much the two overlap as a function of the shift, which is why it captures smoothing and the response of systems.

Why does convolution become multiplication after a transform?

Integral transforms express functions as combinations of exponentials, and convolution acts on each exponential component independently by scaling it. Because the transform separates these components, the combined effect is simple pointwise multiplication in the transform domain.

Convolution

Convolution combines two functions into a third that expresses how the shape of one is modified by the other, the operation at the heart of linear systems and integral transforms.

Definition

The convolution of two functions is the integral, over all shifts, of the product of one function with a reflected and translated copy of the other; it measures the overlap of the two functions as one slides across the other.

Scope

This topic covers the definition of the convolution integral and its discrete analogue, algebraic properties such as commutativity, associativity, and distributivity, the convolution theorem linking it to multiplication under integral transforms, the role of the identity element as a delta function, smoothing by mollifiers, and its appearance as the response of linear time-invariant systems.

Core questions

What does convolving two functions compute?
What algebraic properties does the operation have?
How does the convolution theorem connect it to integral transforms?
Why is convolution the natural model for linear time-invariant systems?

Key theories

Convolution theorem: Under the Fourier or Laplace transform, convolution corresponds to ordinary multiplication, which is why transforms reduce convolution-based problems to algebra.
Linear time-invariant systems: Any linear time-invariant system acts on its input by convolution with its impulse response, so the impulse response fully characterizes the system's behavior.
Approximate identities and smoothing: Convolving a function with a concentrated, integrable kernel smooths it while approaching the original as the kernel sharpens, the basis of mollification and regularization.

Clinical relevance

Convolution models filtering and blurring in signal and image processing, the response of physical systems through their impulse response, probability through the distribution of sums of independent random variables, and the convolutional layers at the core of modern neural networks.

History

The convolution integral appeared in eighteenth- and nineteenth-century work on superposition and in Volterra's integral equations. Its central role crystallized with the operational calculus and the systematic theory of linear systems in the twentieth century, where the convolution theorem made it indispensable.

Key figures

Joseph Fourier
Vito Volterra
Norbert Wiener
Ronald Bracewell

Seminal works

folland1992
bracewell2000

Frequently asked questions

What is an intuitive picture of convolution?: Think of sliding one function across another and, at each position, multiplying them point by point and adding up the result. The output measures how much the two overlap as a function of the shift, which is why it captures smoothing and the response of systems.
Why does convolution become multiplication after a transform?: Integral transforms express functions as combinations of exponentials, and convolution acts on each exponential component independently by scaling it. Because the transform separates these components, the combined effect is simple pointwise multiplication in the transform domain.