High-contrast homogenization: #
Provider.PolynomialHomogenization.Root.EnergyPrice.LawFreeCubeEnergyPrice
Imported from the Apache-2.0 HighContrastHomogenization development at commit
7a13dbcd8d6609264a713373f5c69ceeac870472.
A law-free coarse-grained Dirichlet energy price on one triadic cube #
The attainability route uses the e.cg.RHS mechanism: the energy
of a Dirichlet solution is priced by the boundary datum's fractional norm with
coarse-grained ellipticity constants. This file isolates the exact reason
that price can be law-free.
poincareUpperEllipticityFactor Q a s (.finite 2) is Î_{s,2}(Q;a)^{1/2},
and Î_{s,2} is built only from the coarse-grained block matrix b(R;a) â
a variational response average â never from a pointwise essential supremum of
a. Upstream's
inv_mul_LambdaSq_finite_two_le_card_mul_homogenizationError_sq_add_one
prices it by the identity-referenced multiscale homogenization error. Hence
a single scalar hypothesis, đ †1, converts the whole ellipticity window
into a dimension-only constant, and the resulting Dirichlet energy price
depends on nothing but d and the fractional order s.
The order at which the error is read #
The certificate is a ScalarIdentityPowerTail at
printCertificateOrder g = (1 + g) / 4, while the price needs the
identity-referenced error at the consumer's own order
s â Set.Ico ((1 + g) / 4) (1 / 2). That direction is the available one: the
multiscale error is decreasing in the fractional order, because the geometric weight
concentrates on the coarse scales as s grows, and
homogenizationErrorOnCube_infinity_two_le_of_lt supplies the comparison with loss
exactly one.
Combined with the target reduction â above the common affine scale the retained
amplitude is below correctorTargetAmplitude c Îș = (c / 2) * (1 - 3 ^ (-Îș)), which
is less than one â this yields the single scalar hypothesis đ †1 that the price
below consumes.
The identity reference #
The scalar matrix at 1 is the identity.
Half powers are nonnegative, by the square-root convention.
The multiscale homogenization error at the endpoint spatial exponent and
the finite q = 2 scale exponent is nonnegative.