The Setting
Consider choosing finite families of vectors in as projection directions. Such collections must control the lengths and pairwise independence of the vectors within each family while regulating their averages and extreme inner products against arbitrary test vectors. The question is whether collections meeting these quantitative requirements exist for sufficiently large or under explicit dimension inequalities, and whether every coordinate of every vector can also be kept bounded away from zero.
Direction Families
Letting be a nonnegative integer and a real inner-product space, the empirical covariance of a family at a vector is the real number defined by (1): (1)✦
In a real inner-product space , for any nonnegative integer , the empirical sine transform of a family at is the vector in defined by (2): (2)✦
Given a real inner-product space and a nonnegative integer , the bad projection set associated with a real threshold and a vector is the subset of defined by (3): (3)✦
When is a subset of a set and is a family indexed by a nonnegative integer , the number of bad samples is the nonnegative integer defined by (4): (4)✦
With a nonnegative integer, , and , a direction family in dimension is a collection of vectors in , where with , satisfying the five conditions (5): (5)✦
For a nonnegative integer with and , a pivoted direction family in dimension is a direction family in dimension whose coordinate entries satisfy (6): (6)✦
The Claim
Theorem 1. For every sufficiently large integer , (7)✦ Whenever an integer satisfies (8)✦ a direction family in dimension exists. With the coordinate pivot bound imposed, for all sufficiently large integers , (9)✦
In the paper, the result reads: there is an absolute constant such that, for every sufficiently large integer , with and , there are vectors (, ) such that (1) , within each family the vectors have pairwise nonparallel directions, and every eigenvalue of belongs to ; (2) ; and (3) for every unit vector , at least of the indices satisfy . Whereas the paper uses an unspecified constant and eigenvalues of , the library sets the projection constant to inside and while writing the eigenvalue and sine bounds via and , and it pairs (7) with the explicit dimension claim (8) and the pivoted claim (9).
What Is Checked
The claims (7) and (9) are stated over direction families and pivoted direction families in dimension (built from empirical covariances, empirical sine transforms, bad projection sets, and numbers of bad samples) for sufficiently large nonnegative integers , whereas (8) is stated over direction families in dimension for any nonnegative integer meeting its four inequalities.