The Setting
For a nonnegative integer , let be -dimensional Euclidean space with norm and inner product , and let be the open cube in . For a real-valued function with gradient on a neighborhood of , the question is how the expected displacement between independent uniform draws is bounded in terms of , and how the corresponding inequality behaves under rotations of standard Gaussian vectors in .
Measures and Rotation Paths
The uniform measure on is the probability measure on obtained by restricting Lebesgue measure to the open interval and scaling by .✦ On , the cube uniform measure is the product probability measure of copies of indexed by the coordinates of .✦ The standard Gaussian measure on is the distribution on of a vector with independent standard normal coordinates.
For and , the rotation path is defined by (1)✦ The rotation path tangent at and is defined by (2)✦
The Claim
Theorem 1. For any function , the cube Poincaré inequality (3) states (3)✦ Suppose is and constants and satisfy and for all ; then the rotation Poincaré inequality (4) gives (4)✦ If is -measurable and , then integral rotation invariance (5) reads (5)✦ Whenever is and , the rotation path endpoint bound (6) yields (6)✦
In the paper, the result reads: for a real-valued function on a neighborhood of and independent uniform points , . The bound (3) is stated for on all of under the cube uniform measure and carries the factor on the right-hand side.
What Is Checked
The claims (3), (4), (5), and (6) are stated over -dimensional Euclidean space , the cube uniform measure built from the uniform measure on , the standard Gaussian measure on , and the rotation path and rotation path tangent , for functions (with bounded values and bounded gradient in (4), and a pair in (6)) and -measurable functions at in (5).