The Setting
Let be an integrable real density on the plane, such as a scalar coordinate of the vector measure of a real function of compact support, that vanishes outside the closed disk , and let be its total mass. Localize on dyadic grids of bilinear hats, at scales doubling from a finest scale , and record at each vertex the mass of under its hat. As the scale doubles, positive and negative mass under adjacent fine hats cancel in the coarse coefficient, and the sum of the absolute coefficients can fall below . The question is where that loss sits among the vertices and the levels, and whether each hat piece of can be traded for a density of one sign carrying its coefficient, the difference going into errors of mass zero whose cost is paid by the loss at their vertex.
Dyadic Hats and Defects
The tent function is the piecewise linear bump on of height one at the origin, (1)✦ At a scale , the product hat of an integer index is the real function on given by (2)✦ the piecewise bilinear bump of height one at the point , supported on the closed square of side centered there. Level has the dyadic scale , and for an integer index , called a vertex, is the hat of at that level. At every level, for a natural number , the vertices range over the square box (3)✦ of integer pairs.
The interpolation coefficient of a coarse index and a fine index is (4)✦ which is when , or when differs from by one in exactly one or in both coordinates, and otherwise. The refinement coefficients of depth , for , are the entries of the th power of the matrix : (5)✦
The coefficient of at the vertex of level is its mass under the hat, (6) and the defect is half the absolute mass the coefficient loses against what it merges: at level against under the hat, and at level against the absolute coefficients of level weighted by interpolation, (7)
Roots and Errors
A cancellation scheme over a measure on a space is given by a finite index set , real weights for each level and , and real functions on , the pieces, for . The refined pieces of a cancellation scheme are the real functions on obtained by merging the pieces of the scheme level by level, (8)✦ The root of an integrable real function with respect to is the real function on of constant sign given by if , and otherwise. The roots of a cancellation scheme are the real functions on with the root of the piece and the root of the weighted sum of the roots of level , and the errors of the scheme are the real functions on given by the difference between the function whose root is taken and that root: (9)✦ (10)✦
The root and error densities and of on the box are the roots and errors of the scheme over Lebesgue measure on with index set , the weights at every level, and the pieces of under the hats of the finest level.
The Claim
Theorem 1. Let and , let be a natural number with , and let be an integrable real density on that vanishes outside the closed disk , with total mass , coefficients and defects as in (6) and (7), and root and error densities and on the box . Then for every level and every vertex , (11)✦ the root and the error at satisfy (12)✦ (13)✦ (14)✦ (15)✦ and (16)✦ For every , (17)✦ Moreover, in every cancellation scheme over a measure on a space whose pieces are integrable, at every level and index with , (18)✦
In the paper, the result reads: for every , the set of scalar coordinates , there are absolutely continuous measures , and at all vertices , , supported in , with of constant sign, , , and , the defects, such that satisfies , and , and , with finitely many nonzero terms, for every , , where stands for the set of vertices of level , for the number of levels, and for the hats of the vertices and at their levels, for , for the refinement coefficient of and , and the levels are counted from the coarsest.
What Is Checked
The claim is stated over the tent function and the product hats at the dyadic scales , the box , the interpolation coefficients and the refinement coefficients , the coefficients and defects of , and the refined pieces, roots and errors of cancellation schemes, among them the scheme with the hat pieces and the interpolation weights that gives the root and error densities of .