A binary operation that combines two functions (or sequences) into a third by integrating (or summing) the product of one function with a shifted copy of the other: (f * g)(t) = ∫ f(τ) g(t−τ) dτ (or discrete analogue).

Principle

Translation-invariant linear aggregation: convolution synthesizes the response of a linear, shift-invariant system by superposing shifted, weighted contributions of an input.

Demonstration

Smoothing a noisy time series by convolving it with a narrow, positive kernel reduces high-frequency fluctuations and produces a locally averaged signal.

Misapplication

Treating convolution as pointwise multiplication in the original domain; confusing convolution with simple multiplication or with correlation without accounting for the shift and sign convention.

Consequence

Convolution with suitable kernels yields filtering, smoothing, and linear system responses; repeated convolutions affect spread and regularity of signals or probability densities.

Reversal

Deconvolution seeks to recover an input given an output and a kernel, and is typically ill-posed and sensitive to noise, contrasting with the stabilizing averaging effect of convolution.

Boundary

Defined when integrals or sums converge (e.g., functions in L^1 or sequences summable); not all function pairs admit a well-defined convolution and care is needed with distributions or unbounded domains.

Semantic Tension

Often contrasted with correlation: correlation measures similarity under shift, while convolution inserts a reversal or sign convention in the shift; both are related but serve different inferential roles.

Synthesis

Convolution is the translation-invariant linear combination of one function with shifted copies of another that implements filtering, accumulation, or system response when integrability conditions permit.