High-contrast homogenization: Provider.Response.AnnealedWeakQuantity #
Imported from the Apache-2.0 HighContrastHomogenization development at commit
7a13dbcd8d6609264a713373f5c69ceeac870472.
The annealed weak quantities as integrals of samplewise majorants #
The weak quantities of the profile are squared L² norms of samplewise weak
seminorms. A samplewise estimate for a term of the response split therefore
has to be turned into an estimate for the expectation of that term, and its
samplewise majorant has to be recognized as the integrand whose integral is the
weak quantity.
Both steps are recorded here. The squared L² norm of an extended-nonnegative
observable is the integral of its square, which identifies the weak quantities
with integrals of the squared weak roots; and the expectation of a real
observable is dominated by the integral of its samplewise majorant, which is the
Jensen step. Nothing about the geometry of the weak seminorm is used.
The squared norm as an integral #
The square of the L² norm of an extended-nonnegative observable is the
integral of its square.
The primal weak quantity is the integral of the squared primal weak root.
The coefficient-transpose weak quantity is the integral of the squared adjoint weak root.
The Jensen step #
The expectation against a samplewise majorant. A samplewise bound by a constant multiple of an extended-nonnegative majorant bounds the expectation by the same multiple of the integral of that majorant.
The weak components #
The primal weak component. A samplewise bound by a multiple of the squared primal weak root becomes a bound for the expectation by the same multiple of the primal weak quantity.
The coefficient-transpose weak component.