LocalizationOverlapTransfer #
Ported from scottnarmstrong/Superdiffusion at 7f20476385d4af418f110303085ad0740c9eba46. Module imports, support namespace and local option directives adapted for Lean Pool.
Provider: the bounded-overlap covering transfer for the meso windows #
Source displays in ABK26:
e.nablaw.oscillations(label; display), whose outer window isz + cu_{m-h};e.lower.bound.oscillations(label; display), which averages over the meshz in 3^n Zd cap cu_Kand is deduced "bye.nablaw.in.L.eight" -- a statement about the single global normalized norm oncu_K;e.nablaw.in.L.eight(label; display).
The passage from the global quantity on cu_K to the mesh average of window
quantities is the covering step that the manuscript does not write out. This
module supplies it.
What is proved #
With N = m - h - n the gap and W_R = z + cu_{n+N} the window based at the
site z = triadicCubeShift R of a scale-n cube R:
sum_setIntegral_openCubeAtScale_le-- for a nonnegativefintegrable oncu_K, and a family of scale-ncubes whose windows lie incu_K,sum_R int_{W_R} f <= 3^{dN} int_{cu_K} f. The multiplicity3^{dN}is the exact overlap count ofLocalizationOverlapCount, and is attained there.cubeFamilyAverage_volumeAverage_openCubeAtScale_le_of_card-- the same statement after both normalizations, at a caller-supplied constantCcontrolled by the cardinality gate3^{d(K-n)} <= C |I|. This is the general form: the constant is exactly the ratio of the full lattice count3^{d(K-n)}to the number of cubes actually summed, and the overlap multiplicity has cancelled against the window volume completely.cubeFamilyAverage_volumeAverage_openCubeAtScale_interior_le-- the instance the display consumes, oninteriorMesoCubeGrid d K n (n + N - 1), at the absolute constant3^d.
Where the constant comes from, exactly #
The multiplicity 3^{dN} and the window volume 3^{(n+N)d} cancel exactly:
the volume-normalized transfer constant produced by the covering alone is 1.
The entire constant 3^d of the interior instance is the interior-grid
deficit: the interior grid is only guaranteed to contain the scale-n
descendants of the concentric child cu_{K-1}, i.e. at least 3^{-d} of the
full grid (card_mesoCubeGrid_le_three_pow_mul_card_interior, module
LocalizationGrid), and the grid average divides by the number of cubes
summed. On the full grid mesoCubeGrid d K n the same argument gives the
constant 1, but the full grid's boundary windows are not contained in cu_K
and the hypothesis hsub fails there; that is the content.
Divergences from the printed statement #
- The manuscript's unrestricted mesh is not such a family, and no bound at the boundary sites is claimed here.
- The count is not clipped.
sum_setIntegral_openCubeAtScale_leuses the full lattice multiplicity3^{dN}; clipping the family tocu_Kcould only improve it, and no such improvement is claimed. - Nonnegativity is required everywhere, not almost everywhere. The applications supply a pointwise nonnegative integrand (a squared or fourth power), so the cheaper hypothesis is used. Nothing below is false for the a.e. version; it is simply not proved here.
The covering bound on the sum of window integrals #
The bounded-overlap covering bound. For a nonnegative integrable f and
a family of scale-n cubes whose windows z + cu_{n+N} lie in cu_K, the sum
of the window integrals exceeds the integral over cu_K by at most the overlap
multiplicity 3^{dN} of card_filter_mem_openCubeAtScale_le.
: the caller supplies hI, hsub, hf, hf0.
The normalized transfer #
The normalized bounded-overlap transfer. After dividing by the window
volume and by the number of cubes summed, the overlap multiplicity has cancelled
against the window volume exactly, and the whole constant is the cardinality
ratio supplied by hcard.
: the caller supplies hInon, hI, hsub, hcard, hf, hf0.
The transfer on the interior meso grid, at the absolute constant 3^d.
This is the form e.lower.bound.oscillations consumes. The constant is not
the overlap multiplicity -- that has cancelled exactly -- but the interior-grid
deficit of LocalizationGrid.
: the caller supplies hn : n <= K - 1, hN : n + N <= K - 1, hf, hf0.