Skip to content

Local nonlinear geometry and estimates · Part 10

A Wilson Yang–Mills Construction: A Natural-Language Mathematical Exposition

weights bi, and integer sizes di, with
X
Source page 112.
Searchable transcription
weights bi, and integer sizes di, with
X
i:oi=y
bieAdi ≤γ
for every fine site y.
Source page 112.
Searchable transcription
i:oi=y
bieAdi ≤γ
for every fine site y.

Suppose their coarse supports Xi obey, for x ∈Xi,

dist(x, p(oi)) ≤W(1 + di),
covB(Xi) ≤v + v′di.
Source page 112.
Searchable transcription
dist(x, p(oi)) ≤W(1 + di),
covB(Xi) ≤v + v′di.

Let gi ≤Cbi(1 + di)k, where gi, C ≥0. If a ≥0, ε > 0, and av′ + 2ε ≤A, then

X
giea covB(Xi) ≤eavγCP(2W + 1)4ε−(4+k)(4 + k)! eε
1 −e−ε
e−εm.
(NF9)
x∈Xi
di≥m
Source page 112.
Searchable transcription
X
giea covB(Xi) ≤eavγCP(2W + 1)4ε−(4+k)(4 + k)! eε
1 −e−ε
e−εm.
(NF9)
x∈Xi
di≥m

The sum on the left is over labels i, with x fixed. To prove it, first restrict to di = j. All anchors lie over the coarse ball of radius W(1 + j) about x, so there are at most P(2W + 1)4(1 + j)4 possible fine anchors. Sum the given anchored budget over these anchors and use

giea covB(Xi) ≤bieAdiC(1 + di)keav−2εdi.
Source page 112.
Searchable transcription
giea covB(Xi) ≤bieAdiC(1 + di)keav−2εdi.
The inequality
(1 + j)4+k ≤ε−(4+k)(4 + k)! eεeεj
Source page 112.
Searchable transcription
The inequality
(1 + j)4+k ≤ε−(4+k)(4 + k)! eεeεj

follows by retaining one term in the exponential series at ε(1 + j). The degree-j bound is therefore the numerator in (NF9) times e−εj. Sum its geometric tail. This proof also covers infinitely many address variants, since every finite partial sum has the same bound and all summands are nonnegative.

For a block ratio L, P = L4: a fine site is uniquely specified by its coarse parent and its four residue coordinates. For the halo enlargement of the parent image of an old support, the geometric cover estimate used earlier reads

covB(Xi) ≤v + 10v
L di,
v = (2(l + 6) + 1)4,
Source page 112.
Searchable transcription
covB(Xi) ≤v + 10v
L di,
v = (2(l + 6) + 1)4,

where di = covM(Yi), M ≤2B, and the enlargement has halo radius lB. Each point of it is within (l + 20)B(1 + di) of the parent of any chosen anchor in Yi. Thus (NF9) applies with W = (l + 20)B and v′ = 10v/L. In the inherited sales bound the extra tariffis e2εdi; absorb it first into the old weight, replacing the anchored exponent δ by δ −2ε. The reserve condition is then exactly

a(10v/L) + 2ε ≤δ −2ε.
Source page 112.
Searchable transcription
a(10v/L) + 2ε ≤δ −2ε.

This gives the stated payment, with its L4 child factor and polynomial-degree cost, rather than an unspecified enlargement constant.

The geometric cover estimate just used has the following direct proof. Write L for the block ratio and use the periodic maximum-coordinate distance throughout. A site and the embedded parent of that site are at distance at most L/2. Embedded coarse distances are multiplied by L. The triangle inequality therefore gives

L dist(p(x), p(y)) ≤dist(x, y) + L.
(NF10)
Source page 112.
Searchable transcription
L dist(p(x), p(y)) ≤dist(x, y) + L.
(NF10)

These statements use centered residue representatives and remain valid on winding paths.

Choose a realizing M-cover A of Y , with |A| = d = covM(Y ). Its centers are connected at distance 2M + 1. A unit-step walk visits them all in at most

N ≤d(4M + 2)
Source page 113.
Searchable transcription
N ≤d(4M + 2)

steps. One way to construct it is to start at one center and repeatedly attach an unvisited center adjacent to the visited connected set, inserting an out-and-back detour of length at most 2(2M +1). For the empty cover use the empty-support case separately.

Sample this walk at times LBj, including time zero. Its parent samples form a coarse set D with LB|D| ≤N + LB.

Successive samples, and the parent of any intervening walk point and its preceding sample, are within 2B by (NF10). A fine point of Y is within M of a visited cover center, so its parent is within 3B of the parent of that center, since M ≤2B and B ≥1. It follows that the halo of radius lB around p(Y ) lies within (l + 5)B of D, with the required connected covering centers.

To replace this larger-radius cover by a B-cover, tile each such ball with points spaced B apart in the four coordinate directions. For a ball of radius mB, the integer coordinate box with |zi| ≤m+1 suffices: coordinatewise integer division puts every point within B of one of its grid points. The grids are connected, and the connections between their centers join their union. Their combined cardinality is at most |D|(2(m+1)+1)4; taking images on a torus only decreases cardinality. With m = l + 5, this gives covB(halolB p(Y )) ≤|D|(2(l + 6) + 1)4.

Finally, 4M + 2 ≤10B. Multiply the last inequality by L and use the sampled-walk bound to obtain L covB(halolB p(Y )) ≤(L + 10d)(2(l + 6) + 1)4,

as claimed. The same walk bound gives dist(x, y) ≤d(4M + 2) + 2M for x, y ∈Y . Combining it with (NF10) and the halo radius proves the anchor-distance estimate with W = (l + 20)B. This completes the geometric input to (NF9).

Source: A Wilson Yang–Mills Construction: A Natural-Language Mathematical Exposition. Both papers belong to one Zenodo record. No external funding.