Distributions 1 | Motivation and Delta Function [dark version]

Distributions were introduced to make sense of derivatives and other operations that break down at “sharp” features—especially jumps like the Heaviside step function. In classical calculus, the derivative of the Heaviside function H(x) (0 for x<0 and 1 for x>0) fails to exist at x=0 because of the jump. Yet many applications—most notably differential equations and Fourier analysis—need a derivative-like object that still behaves correctly even when solutions have discontinuities or corners.

The motivation traces back to Paul Dirac’s 1927 work. Dirac wanted a new “delta function” δ(x) that would act as the derivative of the Heaviside function. Away from x=0, this is straightforward: H is constant on each side, so any derivative-like object should be zero there. The real challenge is the single point at x=0. Dirac also demanded that a fundamental theorem of calculus–type identity remain valid in an integral form: integrating δ(x) over a tiny interval around 0 should reproduce the jump in H. Concretely, for arbitrarily small ε, the integral of δ(x) from −ε to ε should equal H(ε)−H(−ε), which is 1.

But those requirements clash with what ordinary functions can do. If δ(x) were an ordinary function that is zero everywhere except at one point, then its integral over any interval would be zero—because changing a function at a single point does not affect integrals (in the sense of “almost everywhere,” i.e., with respect to the Lebesgue measure). That produces a contradiction: the integral would have to be 0, not 1. The situation worsens when one tries to differentiate δ again; even writing down δ′ requires a framework that can interpret such expressions without relying on pointwise values.

The response is to replace “functions” with a broader concept: distributions, also called generalized functions. Instead of treating δ as a literal function with a graph, distributions are treated as linear functionals acting on test functions. The guiding picture is that distributions behave like mass densities. For an ordinary function f(x), integrating f(x) against a test function φ(x) over a region produces a real number that can be interpreted as “mass measured” in the bump-shaped region where φ is nonzero. For the delta distribution, all the mass is concentrated at x=0. When a test function φ is used, the resulting number corresponds to φ evaluated at the point of concentration—so δ “picks out” the value at 0.

This shift—from pointwise objects to measurement-by-integration—lets δ satisfy the integral jump condition and supports further algebraic operations like differentiation in a consistent way. The next step is to formalize what counts as a test function and to define precisely how distributions act as linear maps on them.