A binary operation on functions, measures, or distributions defined (in the classical case) by (f * g)(x) = ∫ f(y) g(x − y) dy (or an analogous formula on groups) that produces a combined function used for smoothing, filtering, and representing translation-invariant linear operators.

Principle

Convolution averages one function against translates of another; it is commutative and associative on abelian groups, corresponds to pointwise multiplication under the Fourier transform, and realizes translation-invariant linear operators as convolution by a kernel.

Demonstration

Convolving a function with a mollifier yields a family of smooth approximations f * ρ_ε that converge to f in suitable norms; in signal processing a filter is implemented by convolution with an impulse response; in PDEs, convolution with a Green's function gives particular solutions.

Misapplication

Formally computing f * g without verifying integrability, support, or distributional pairing conditions can be invalid; assuming convolution is commutative on non-abelian groups or neglecting boundary conditions on domains with edges leads to incorrect results.

Consequence

Provides a flexible toolkit for regularization, construction of approximate identities, analysis of linear time-invariant systems, and conversion of differential/integral operators into multiplicative forms in the frequency domain facilitating estimates and inversion (when possible).

Reversal

The dual operation is pointwise multiplication in the original domain corresponding to convolution in frequency; deconvolution (inverting convolution) is often ill-posed and amplifies noise, reversing smoothing into instability.

Boundary

Definition and basic algebraic properties depend on the domain (Euclidean space, locally compact group), the function spaces involved (L^1, L^2, tempered distributions), and on group commutativity; in non-abelian settings convolution is generally noncommutative and requires Haar measure choices.

Semantic Tension

Tension between convolution as an algebraic product versus as an integral transform: in analysis one emphasizes mapping and regularization properties, while in algebraic or representation-theoretic contexts one treats convolution as an algebraic multiplication on function spaces or group algebras.

Synthesis

Convolution is the integral averaging of one function against translates of another that implements smoothing and encodes translation-invariant operators; via the Fourier transform it becomes pointwise multiplication, unifying time/space-domain averaging with frequency-domain algebra.