Keyboard shortcuts

Press ← or → to navigate between chapters

Press ? to show this help

Press Esc to hide this help

Gaussian-weighted integral operators

Abstract

Gaussian density gives the actual integral-kernel factory, density isometry, difference-kernel estimates, and convolution conjugacy.

Theorem 1.1 (The actual L2 kernel factory).

Proof. Machine-checked in Lean as D5/S3/Fourier/Asymptotics/CosineCutoffKernel/GaussianWeight.weighted_integral_operator (✓ std3). ∎

Source. Repository-derived.

Commentary.

For the original parameters 0<r<1 and arbitrary real alpha, set a=(1+r)/2, b=(1-r)/2, c0=1/(4pisqrt(ab)), kappa=1/a+alpha^2/b, rho(x)=c0exp(-kappa*x^2/2), and mu=volume.withDensity(ofReal(rho)). The displayed T is a continuous linear map from real L2(mu times mu) to continuous operators on real L2(mu). Its representatives are the actual kernel integrals, it is 1-Lipschitz in the kernel, it preserves arbitrary filter limits, and its raw-representative clause proves row integrability and equality for any almost-everywhere-equal representatives. An eventual uniform bound passes to the limit only under the displayed nonempty-filter hypothesis.

Theorem 1.2 (An onto density isometry).

Proof. Machine-checked in Lean as D5/S3/Fourier/Asymptotics/CosineCutoffKernel/GaussianWeight.density_unitary (✓ std3). ∎

Source. Repository-derived.

Commentary.

With exactly the same density and measure, the real linear isometry equivalence U multiplies a weighted-L2 representative by sqrt(rho) almost everywhere for Lebesgue measure. Its inverse divides a Lebesgue-L2 representative by sqrt(rho) almost everywhere for mu. The strictly positive Gaussian density makes both directions lawful; surjectivity and the inverse formula are included in the equivalence, not assumed separately.

Theorem 1.3 (Gaussian difference kernels and their limits).

Proof. Machine-checked in Lean as D5/S3/Fourier/Asymptotics/CosineCutoffKernel/GaussianWeight.gaussian_weighted_kernel (✓ std3). ∎

Source. Repository-derived.

Commentary.

Put g0=c0^2sqrt(pi/kappa). The Gaussian overlap integral is g0exp(-kappa*z^2/4), including its exact z=0 square-density identity. Every measurable real L2 function q supplies an actual weighted product-L2 kernel represented by q(x-y), with the displayed exact norm-square integral and norm bound. Every pair of such actual kernels obeys the difference estimate. For any index type and filter, convergence of the actual unweighted L2 classes implies convergence of every represented weighted kernel family. All representative equalities, measurability assumptions and MemLp witnesses are quantified explicitly; the family limit does not require a nonempty filter.

Theorem 1.4 (Conjugacy of the same integral operators).

Proof. Machine-checked in Lean as D5/S3/Fourier/Asymptotics/CosineCutoffKernel/GaussianWeight.cutoff_conjugacy (✓ std3). ∎

Source. Repository-derived.

Commentary.

The same real function q supplies both the unweighted operator C and weighted operator T through their actual ordinary integral representative equations. The equivalence U has the displayed density and inverse-density formulas. Multiplication by sqrt(rho) defines M with norm at most sqrt(c0); the exact equation UTU-inverse=MCM yields norm(T) at most c0*norm(C). This supplier assumes the two displayed integral representations for its arbitrary input operators. The original finite-cutoff client separately constructs those actual operators before applying it; no assumed representation replaces that client construction.

References