The Setting
For a centered score of unit second moment on a finite alphabet, the question is whether deterministic tests of finite labels can retain a fixed positive quadratic deficit. The labels may be observations of the score through a prescribed map.
For a finite alphabet and real functions , the uniform mean is the average and the normalized sum is the function defined in (1); the mean on the empty alphabet is zero. (1) All expectations below are uniform on the indicated finite product alphabet. For a finite leaf set and real weights , the soft event weight is the function in (2). (2)✦ For a score , a real center and a real weight , the quadratic deficit is the weighted average in (3). (3)✦
The Fixed Gaussian Tests
The Gaussian construction uses the grid half-size , radius , spacing , tent width and leaf set in (4); its middle index is . (4) Clipping is a function from to . Write for the indicator of an event . The grid ramp difference , selector partition weight , leaf center and leaf tent test are the functions and numbers in (5); denotes a tent with center and width parameter . (5)
For a measure on a measurable space , a real score and a family of real functions on , the weighted moment array is the real array of orders defined in (6). For a finite alphabet , a real score and real weights , the finite weighted moment array is the real array defined in (6). (6) The weighted means and variances of a moment array are the ratios in (7). (7) For real moment arrays of orders zero, one and two, the moment gain is the real number defined by (8). (8) Writing for the standard Gaussian law, the Gaussian leaf and selector moment arrays and the gain threshold are those of (9). (9) The finite tests and their centering value for the normalized score in (1) are denoted by (10). (10)
Deterministic Tables
For finite sets , a selector table is a map from to , and a leaf test table is a family of maps , with meaning that the test passes. The unique failing leaf event associated with a selector table and a leaf test table on is the subset of defined in (11). (11) For real weights on a finite set , the product weight is the real function on defined in (12). The Bernoulli weight for a real number is a real function on , and the leaf table weight for a real function is a real function on tables ; both are defined in (12). (12) For real arrays , the rounding table weight on pairs is the product in (13). (13)✦ A label map acts on all selector and leaf coordinates through ; its coordinatewise map on length samples is , as in (14). Write and . (14)
The Claim
Theorem 1. Let be finite and nonempty, and let have uniform mean zero and second moment one. There is a positive integer with the soft gain (15): (15)✦ For finite and real arrays , suppose that for every . The table averages satisfy (16) for every , and (17) for every map , real score and center : (16)✦ (17)✦ Suppose in addition that and for all . For any real , there are tables with the deterministic deficit (18): (18)✦ Finally, let be a map to a finite label alphabet and let satisfy and . With formed from , there are a positive integer , a real , and tables and with the gain (19): (19)✦
In the paper, the result reads: the Gaussian event and predictor are fixed before transfer, normalized finite scores have an eventual positive deficit, and the cell cost and retained variance bounds are then applied to amplify.
What Is Checked
The soft and final table gain statements concern centered scores of second moment one on uniform finite nonempty alphabets and their product samples; the averaging identities concern real selector and leaf arrays on finite label sets, with selector normalization, and deterministic extraction also assumes nonnegative selectors and leaf weights in .