distribution (generalized function)

Physicists long wrote down objects like the Dirac delta — an ‘infinite spike at one point with total area one’ — and the derivative of a step function, which classical analysis says do not exist. Distributions are the rigorous home for such objects. The key shift in viewpoint: stop asking what an object's value is at each point, and instead ask only what it does when integrated against a smooth test function. A distribution is defined entirely by those weighted averages.

Formally, fix the space of test functions: smooth functions with compact support (or the Schwartz space of rapidly decreasing functions, for tempered distributions). A distribution is a continuous linear functional on this space — a rule T that assigns to each test function phi a number T(phi), linearly and continuously. Every locally integrable function f gives a distribution by phi maps to integral of f phi, so ordinary functions embed faithfully; but there are far more distributions than functions. The Dirac delta is the functional phi maps to phi(0).

The power comes from operations that extend by ‘moving the work onto the test function’. The derivative of a distribution T is defined by T'(phi) = -T(phi'), copying integration by parts — so every distribution is infinitely differentiable, and a jump's derivative is a delta. The Fourier transform extends to tempered distributions the same way, making sense of the transform of objects like 1, e^{i x}, or the delta itself, none of which is integrable. This is the natural setting for partial differential equations and for harmonic analysis beyond classical function spaces.