High-contrast homogenization: #
Provider.PolynomialHomogenization.Root.Dirichlet.OutputFactorEccentricityBound
Imported from the Apache-2.0 HighContrastHomogenization development at commit
7a13dbcd8d6609264a713373f5c69ceeac870472.
(β): the output factor, and the correction to classified #
outputFactor = √(hsAffineFactor (matSqrt (symmPart abar))⁻¹ s ‖matSqrt (symmPart abar)‖)
as "none — it CANCELS". corrected that: the factor cancels on the
left of the terminal bound, but it also occurs inside hRate's left-hand
side, where that module carries it into its own C₀. A law-free C₀ therefore needs
it bounded.
It cannot be bounded by a law-free constant times an eccentricity power
alone. hsAffineFactor is homogeneous of degree s in the absolute scale of
symmPart abar, while the witness eccentricity is scale invariant: at
abar = t • 1 the eccentricity is 1 for every t > 0 while the factor is
t ^ s. The dispatch's (β) is false as literally stated.
The correct form, proved here. The frozen inner-ellipsoid normalization ties the gauge
cube side to that same absolute scale from both sides
(witnessCubeScale_two_sided), and the ball sandwich's inner radius
ρ is below the gauge scale. So the cube side is bounded below by 2ρ/(3√d),
and dividing the energy's paired estimate
hsAffineFactor L⁻¹ s ‖L‖ * ell ^ (2 s) ≤ gaugeFrameConstant d s * ecc ^ (d + 2 s)
by that lower bound leaves a head depending on (d, s, ρ) only — which is
exactly C₀ s₀ ρ Rad's permitted dependence — times ecc ^ (d + 2 s). The
s-dependence of the eccentricity exponent is then trimmed away by s < 1/2
and the monotonicity of the fold factor, leaving the (d)-level exponent
(d + 1) / 2 that the pEcc slot can carry.
The law-free head #
The output factor's law-free head.
Equations
Instances For
The bound #
(β), the corrected form. The affine distortion factor of the gauge
frame change is below a (d, s, ρ)-level head times the (d + 2 s)-power of
the witness eccentricity. The inner sandwich radius ρ is what pays the
absolute scale that the eccentricity cannot see.
(β) in the fold slot. The output factor itself is below the law-free
head times the fold factor at the (d)-level exponent (d + 1) / 2; the
s-dependence of the exponent is trimmed by s < 1/2, exactly as that module trims the
geometric factor's order dependence.