The Setting
Take real scale parameters and a real point . Integrating a signed Gaussian-weighted kernel between the squared scales and produces a smooth kernel, and the question is how it splits into a constant multiple of a hard-truncated Hilbert kernel on and a difference of endpoint errors at scales and , each bounded by a multiple of a Gaussian density.
Kernels and Error Profiles
For each real variance , the Gaussian density is the real-valued function defined for by (1): (1)✦ The signed kernel is the function of two real variables given by (2): (2)✦ Given real scale endpoints and , the smooth kernel is the real-valued function defined for by (3): (3)✦ For real scale parameters and , the smooth kernel over squared scale endpoints is the function defined for by .
The heat kernel constant is the real number .✦ Define the smooth kernel profile as the real function given at by (4): (4)✦ Write for the kernel error profile, the function on satisfying (5) for each : (5)✦
For real scale parameters and , the hard-truncated Hilbert kernel is the real function given for by (6) with : (6)✦ At any real scale parameter , the endpoint error is the real-valued function satisfying (7) for , with : (7) The error envelope constant is a chosen positive real number such that for all .✦
The Claim
Theorem 1. If satisfy , then the smooth kernel decomposes according to (8): (8)✦ Moreover, if satisfy , then the endpoint error satisfies the Gaussian bound (9): (9)✦
In the paper, the result reads: there are an absolute constant and functions () such that, for , apart from endpoint values, with , and for , where with stands for .
What Is Checked
The decomposition is stated for real scales and , and the endpoint bound for real and , over the smooth kernel built from the signed kernel and Gaussian densities , the heat kernel constant , the hard-truncated Hilbert kernel , the endpoint error built from the smooth kernel profile and the kernel error profile , and the error envelope constant . The library also proves the Gaussian bound for the error profile: there exists a positive real constant such that for all , .✦