Fourier inversion theorem

Statement and common formulations

There are several rigorous formulations depending on the function class. For sufficiently nice functions (for example, rapidly decreasing smooth functions known as Schwartz functions), one common normalization of the inversion formula is

f(x) = \int_{\mathbb{R}^n} \hat f(\xi) e^{2\pi i x\cdot\xi}\,d\xi,

where \hat f denotes the Fourier transform. Different authors use alternate constants (such as 1/(2\pi) factors); the precise factor depends on the chosen convention. The theorem also holds in weaker senses: if f lies in L1 and its transform is in L1, the integral gives f almost everywhere; in L2 one has an L2 inversion provided by Plancherel's theorem; and for distributions the inversion holds in the distributional sense. For practical intuition about frequencies and phases see frequency and phase.

Conditions and technical points

• L1 theory: If f\in L1(R^n) and \hat f\in L1(R^n), then the inversion integral exists pointwise almost everywhere and recovers f.
• L2 / Plancherel: For square-integrable functions inversion is guaranteed in the L2 sense; pointwise values may require additional regularity.
• Tempered distributions: The transform and inversion extend to distributions, allowing recovery in a weak sense for generalized functions.

History and development

The idea of decomposing functions into sinusoidal components dates to Joseph Fourier's work on heat conduction in the early 19th century. Mathematical foundations were refined in the late 19th and 20th centuries as concepts of integrability and convergence were formalized. The inversion theorem emerged as analysts clarified conditions under which transform pairs exist and are unique; later work by Plancherel, Riemann, Lebesgue and Schwartz expanded the scope to L2 and distributional settings.

Uses and examples

The inversion theorem underpins many applications. In signal processing it guarantees exact reconstruction of band-limited signals from their spectrum. In partial differential equations, solving linear constant-coefficient equations often reduces to multiplication in the Fourier domain and inversion back to the solution. In probability and statistics characteristic functions (Fourier transforms of measures) are inverted to recover distributions. See concrete introductions at function spaces and applied treatments at wave applications.

Remarks and notable facts

Important practical points include the sensitivity to normalization conventions and to regularity: pointwise inversion may fail at discontinuities (yielding averaged values). The Riemann–Lebesgue lemma ensures that transforms of integrable functions vanish at infinity, while Paley–Wiener theorems characterize supports and analyticity properties via the transform. For further reading and technical proofs consult advanced treatments and survey articles linked from phase resources and frequency discussions.