Fourier transform on Schwartz functions

A function belongs to $\mathcal S(\mathbb R^d)$ when it is smooth and every quantity $\sup_x|x^\alpha\partial^\beta f(x)|$ is finite for multiindices $\alpha,\beta$. This controls not only the function's tail but all derivative tails. Compactly supported smooth functions and Gaussians belong to the class. The function $1/(1+x^2)$ is smooth and integrable on the line but not Schwartz, because sufficiently large polynomial weights overwhelm its decay. Smoothness alone and rapid decay of the function alone are each weaker requirements.

These weighted bounds imply absolute integrability of all derivatives multiplied by any fixed polynomial. They allow differentiation under the Fourier integral and repeated integration by parts, with boundary terms vanishing. They also supply a natural topology: convergence in Schwartz space means convergence of every weighted-derivative supremum, not merely pointwise convergence. This stronger structure will later let distributions act continuously on test functions and makes the transform a continuous invertible map of the same test-function space.

Integrating a spatial derivative by parts moves the derivative onto the exponential, producing multiplication by $2\pi i\xi_j$. Differentiating with respect to frequency instead brings down $-2\pi i x_j$. Repeating these identities bounds every weighted frequency derivative in terms of integrable weighted spatial derivatives. Consequently the Fourier transform maps Schwartz space to itself. The signs are not cosmetic: they determine translation direction and the correct multiplier for a differential operator.

A spatial translation multiplies the transform by a phase, while multiplication of the function by an oscillation shifts its transform. Scaling trades spatial width for frequency width with a volume prefactor. These rules explain how localized oscillatory signals can be analyzed without redoing every integral. They must be applied consistently with the chosen exponential sign and frequency normalization; a formula written for angular frequency has different factors of two pi. Verify one simple Gaussian or translated test function whenever conventions are changed.

In one dimension, the Gaussian $e^{-\pi x^2}$ transforms to itself. One derivation differentiates its transform, uses integration by parts to obtain a first-order differential equation, and fixes the constant with the Gaussian integral. Scaling then gives the transform of $e^{-\pi a x^2}$ for positive $a$. The result has prefactor $a^{-1/2}$ and exponent $-\pi\xi^2/a$. Concentrating the spatial Gaussian therefore broadens its frequency profile while changing its total mass.

The Schwartz calculation is a safe starting point, not a claim that every signal decays rapidly. Indicator functions, power-law tails, and point masses require extensions or different theorems. A differential identity derived in the Schwartz class can later extend by continuity to a larger space or hold in a distributional sense, but that extension must be named. Formal manipulation of a divergent integral is not justified by resemblance to the Gaussian computation. Retain the test class until a separate norm or duality argument supplies the broader meaning.