with 0 the fixed penalty, and likewise its bill norm is at most
). All the face roots lie within
, by the parent-distance bound. Thus
box(
PlaqAxes
dir
(NF2)
No global pure-gauge representation of the background has been assumed.
The clean-background flattening estimate supplies the gauge in (NF1) with), uniformly over the actual clean boxes, where
) has rate
in the source’s rate calculus. The local chord chart supplies this bound by writing each transformed link as the exponential of a Lie-algebra element of norm at most
), then using its uniform chord bound. The chart radius is fixed before taking
large. Put
Searchable transcription
M(x) = 2UCcG(B(x) + 1)Y s(x).
Equation (NF2), and the trivial zero bound on inactive bonds, give
Searchable transcription
∥uhot(e)∥≤URaddr(e)(u) + M(x)Iaddr(e)(u).
The neighborhood census bounds the sums of the squared local quantities by their global sums times swept() = ballCost(
Axis
Searchable transcription
Squaring the preceding inequality with (a + b)2 ≤ 2a2 + 2b2, summing, and requiring 2M(x)2 swept(L, 3B(x)) ≤1/2, yields X e ∥uhot(e)∥2 ≤4U2 swept(L, Rmac + 2) X p ∥motion(u)p∥2. (NF3)
Searchable transcription
This smallness holds on a common tail: the radius census has rate zero, while M has rate −1/2, so the absorbed quantity has negative rate. All constants are chosen after the block side L, before the sector and the variation. Finally the real plate is the motion Gram matrix plus the identity on inactive coordinates. The exact normalization is ⟨v, formT form v⟩= scale2 X p ∥motion(u)p∥2.
The Euclidean square norm splits into its active and inactive squares. If 0 is a common constant in (NF3), the plate therefore obeys
Searchable transcription
⟨v, plate v⟩≥min(scale2 /a, 1)∥v∥2.
This derives the uniform Gaussian floor from one accepted branch’s local forest inequality and clean-background flattening, rather than assuming a floor for the stacked action.
The forest inequality used in (NF2) is obtained directly from the nonlinear slice-shell inequality. For a logarithmic direction that vanishes on the forest trunks, consider
= represent(
). Then
remains on the same forest slice for every real
, and
= 1. All predictor frame weights are usable near
= 0: the minimum condition holds at the identity and on a neighborhood, and there are finitely many coarse roots. The shell inequality, with comparison background the identity, gives
chord( port
) + chord(prediction(
Each local port is bounded by the full sum of plaquette chords over the macro box. This may overcount periodic plaquettes, which only enlarges a nonnegative sum.
If a group-valued curve starts at the identity and has matrix derivative colorMat(
), the normalization of chord and the color norm imply
Searchable transcription
|t|−1 chord(Vt) −→∥v∥.
Indeed the derivative quotient of its matrix is colorMat(); take its norm and divide by the fixed color scale. Apply this fact to the edge, to each plaquette holonomy, and to the actual predictor. Their derivatives are respectively the collected edge variation, its linear curl, and the predictor bill. Divide the shell inequality by
, let
0 through nonzero real values, and pass through the finite port sums. The resulting inequality is exactly the forest inequality used in (NF2). In particular its pin term is the derivative of the nonlinear predictor chosen by the construction.
Searchable transcription
This argument also gives the fixed-block vertical Hessian gap. If X = box(Rmac), squaring the forest inequality and applying Cauchy–Schwarz to the two finite sums yields X X X ∥w(e)∥2 ≤β ∥curl w(b)∥2 + γ ∥bill w(b)∥2, e b b
Searchable transcription
with β = 2K2L4| Axis | |X|2| PlaqAxes |, γ = 2L4.
For fixed offset the map from coarse root and plaquette type to the fine plaquette is injective, so summation over coarse roots costs, even when different offsets represent the same periodic site. Parent fibers have exactly
elements. These two observations account for all factors and avoid any period-dependent norm equivalence.
On a vertical forest coordinate, the stretched field has collected value zero on trunks and
on the complementary bonds. The Hessian of the tested action is bounded below by
Searchable transcription
a X b ∥curl w(b)∥2 + k X b ∥bill w(b)∥2,
where 0 is the curvature-metric floor. The pin term has this sign because the vertical coarse mismatch is minus the bill. Therefore
Searchable transcription
X Hess(v, v) ≥min(a, k) e/∈trunks ∥ve∥2. (NF4) 1 + β + γ
The Hessian restriction follows by composing the joint analytic action with the linear vertical inclusion; its second derivative is the joint second derivative evaluated on two vertical vectors.
The local stencil comparison in (NF1) has a separate elementary transport proof. A carrier from to
has path length at most 4 dist(
). When all links on a surrounding radius-
box have chord at most
, and that path lies in the box, the carrier’s chord is at most
, by the product chord inequality. To apply it using only local hypotheses, replace all links outside the box by the identity; locality leaves the carrier unchanged. The matrix adjoint estimate
Searchable transcription
∥Adg b −b∥≤2cG chord(g)∥b∥
then bounds straightening of a local variation of size. A bounded stencil of norm at most
, depending only on its declared lookup neighborhood, transfers this to a row error at most
. One can see the locality step explicitly by cutting the difference of the two variations to that neighborhood; its sup norm is at most the local bound, and the stencil sees the same difference.
Under a signed coordinate change, a reversed link contributes a further adjoint correction of at most 2; forward links contribute none. The transformed lookup is contained in the original box enlarged by one. The total comparison is thus
Searchable transcription
∥orientedRow(V, u) −orientedRow(1, u)∥≤2CcG(d + 1)hM.
Taking the box and variation bounds specified before (NF1) proves the asserted estimate. Adjoint action and the signed displacement preserve each individual variation norm.
The same transport calculation preserves the entrywise decay needed for the matrix upper price. A single-bond impulse remains a single impulse after the signed displacement and carrier conjugation, and its color norm remains unchanged. The untransported stencil’s impulse decay therefore implies
Searchable transcription
ew dist(e,s focus(p))∥orientedRows,p(V, 1eb)∥≤2Cew(predReach +7)∥b∥,
where 0 is the common stencil width. The extra one in this exponent pays the possible signedlink endpoint shift. Thus the local comparison does not introduce a volume factor into the matrix estimate. Finally the small logarithmic background in (NF1) is constructed, not assumed. For a clean radius-(
+ 1) box,
), axial gauge flattening gives a gauge
with chord at most
Searchable transcription
u(x) = 16(2(32B(x) + 1) + 1)5| PlaqAxes | 2 fiery(x)
on the radius- box. The accepted temperature comparison bounds the clean plaquette threshold by 2 fiery(
). A fixed local logarithm chart extends these transformed links to a whole logarithmic array
, agreeing on this box, with
), once
) lies in its fixed chart radius. The radius factors have rate zero and the fiery threshold has rate
2; hence
) has rate
2 and eventually lies in that chart. This supplies the function
) used in (NF3).
The logarithmic product estimates used here follow from the analytic local map log(exp
exp
). It satisfies
0) = 0 and
, by differentiation of matrix multiplication and the local inverse relation between logarithm and exponential. A bounded second derivative on a sufficiently small closed ball therefore gives
Searchable transcription
∥B(a, b) −(a + b)∥≤C∥(a, b)∥2, ∥B(a, b)∥≤C∥(a, b)∥.
The local exponential recovers the product on that ball. Applying this at each link gives the same constants for every finite lattice in the sup norm, since the operation is coordinatewise. This justifies use of a genuine left group chart in the comparison, including its quadratic error.
The extension of local logarithms used above has a particularly simple form. The analytic matrix logarithm chart satisfies log 1 = 0 and, on a fixed neighborhood,
Searchable transcription
exp(log U) = U, ∥log U∥≤C chord(U).
The second inequality follows from the local Lipschitz bound of the analytic chart andmat(
chord(
). Choose the chord radius small enough to lie both in the Lipschitz neighborhood and in the neighborhood of the right-inverse identity. On the requested local set of edges define
= log
, and outside it define
This is a whole finite array with
; its exponential agrees with
on the requested set. No continuous cutoffor global logarithm of
is involved.
The adjoint estimate in (NF1) is likewise a direct matrix inequality. For a unitary and a color matrix
Unitary invariance of the Frobenius norm and submultiplicativity bound its norm by 2 Dividing by the color scale proves the stated physical-color inequality. The path chord inequality follows by induction on path length, using the product triangle inequality and invariance of chord under inversion; it also applies to paths with repeated edges. A related estimate needed in the local vertex remainders is
Searchable transcription
∥curvature(face(exp b)) −curl b∥≤C∥b∥2 ∞
on a fixed sup-norm ball, with the same constant for every lattice. To prove it consider the analytic finite-dimensional map
curvature(exp
exp
exp(
) exp(
It vanishes at zero and its derivative there is. Taylor’s theorem with bounded second derivative gives the quadratic error and a uniform bound on this linear part. Substituting the four actual face coordinates gives the result; their maximum norm is at most
, even when periodicity identifies two or more of these coordinates.
Common envelopes for the non-packet activity letters
The regular-log and forced-letter estimates must use one envelope independent of the background sector, its forest, and the Gaussian variable. The assembly is as follows. Fix 0 below one quarter of the support exponent and below one eighth of each of the two stencil widths, a nonnegative footprint weight
, and a retained order
4. Choose the block side large enough for the inherited footprint estimates. After this choice fix the positive Gaussian floor and matrix upperprice parameters, the local chart constants, and the remaining real sector parameters. Suppose the covariance-window error has rate
For every exponent 0 and any prescribed forced-letter exponent
, there are common nonnegative weights
for forced letters and
for the other logarithmic letters such that
Searchable transcription
| bagi(U)| ≤wi, | farLogp(z)| ≤yp, |efarLogp(z) −1| ≤yp. (NF5)
Searchable transcription
Their weighted footprint sums have respectively budgets βu F and βb F . These are the sales inequalities used in the connected expansion: the weights include the specified footprint exponential, overlap price, radius charge and retained derivatives. All the ancillary costs enter before the budget is reduced to the displayed powers. To see the exponent bookkeeping, the local estimates give a base envelope b(0) p , obtained by combining the inherited physical letters, the finite local logarithmic letters and the covariance- window letters. Their combined sales budget has the form
Searchable transcription
T(x) = βF (x)s + X(x) + N(x), −1/2 < s < b,
Searchable transcription
where X, N have rate −1/2. Thus T has rate s; eventually T ≤1 and 2T ≤βb F . The envelope is chosen by evaluating the uniform estimates on an empty-seed dummy sector at the same radius and cutoff, then retaining their background-independent weights. This is legitimate because the sector hypotheses hold there and the estimates themselves quantify over all real backgrounds. It makes no choice based on the subsequently integrated Gaussian value.
The base envelope bounds each logarithm and is individually at most one, since a nonnegative weighted sales sum bounds each of its summands. For1, the elementary bound
supplies the Mayer envelope. Consequently set
; multiplying its sales budget by two costs at most
, as just arranged. The forced-letter envelope keeps its separately prescribed budget. The real localization factors all lie in [0
1], so multiplying by them can only decrease these pointwise bounds. This establishes (NF5) with one set of weights usable in every sector and for both the raw logarithms and their Mayer factors.
Searchable transcription
The letters combined in (NF5) are the following actual terms. Besides the inherited activity logarithms, there are three finite local types: row-energy remainders indexed by a signed orientation and a row, a diagonal logarithmic correction indexed by a fine bond, and a Haar-chart correction indexed by an active fine bond. There are also five chain types: the principal log chains for the real plate with coefficient −1/2, the corresponding blank-plate chains with coefficient +1/2, the covariance correction chains with coefficient +1/2, the affine first-order correction, and the residual quadratic chain. Their chain lengths are their additional charges; the affine bond term has charge zero. Their supports are the corresponding full receipts enlarged by the halo parameter lB, including every lookup used by the term.
For a row with coarse center, the actual short exponent is zero within distance 200
of the seeds, and otherwise is
Searchable transcription
−βθ [go(u) lighto(u) + extractedo(μ)] , 0 ≤go(u) ≤1, 0 ≤θ ≤1. (NF6)