Coarse-graining support: Support.HighContrast.EntryScale.DeterministicAlgebra.P2 #
Imported from the Apache-2.0 CoarseGraining development at commit
c7ddd76c08ade64fed1b8d2ca51be14dfee8deb4.
Source label e.drift.general: if both scalar ratios are at least one, then
each diagonal drift entry is bounded by the product excess a b - 1.
Euclidean/L2 operator norm of a full block matrix. This is the matrix norm
used by the full Ahom_m^{-1/2} (...) Ahom_m^{-1/2} observable in source
label l.union.bound.
Instances For
Source label l.S.and.J: squared triangle inequality for the full-block
operator norm, in the form used to split
A(cu_j) - Ahom_m = (A(cu_j) - Ahom_j) + (Ahom_j - Ahom_m).
Source label l.union.bound: full-block operator-norm bridge for changing
both sides of the terminal normalization. Once the two diagonal change
matrices have norm bounds, this converts them into the corresponding bound for
the full centered block observable.
Source label l.union.bound: diagonal entries of the full-block
terminal/intermediate normalization-change matrix, written in terms of the two
scalar ratios
barSigma_j / barSigma_m and
barSigmaStar_j^{-1} / barSigmaStar_m^{-1}.
Equations
- HCPolySupport.HighContrast.EntryScale.terminalNormalizerChangeDiag upperRatio invStarRatio (Sum.inl val) = √upperRatio
- HCPolySupport.HighContrast.EntryScale.terminalNormalizerChangeDiag upperRatio invStarRatio (Sum.inr val) = √invStarRatio
Instances For
Source label l.union.bound: concrete diagonal entries of
Ahom_m^{-1/2} Ahom_j^{1/2} in the scalar-block coordinates.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Source label l.union.bound: concrete full-block diagonal matrix for
Ahom_m^{-1/2} Ahom_j^{1/2}.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Source label l.union.bound: the library's scalar full-block normalizer at one scale,
as the diagonal matrix Ahom_n^{-1/2}.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Source label l.union.bound: full-block matrix centered by the scalar
annealed block at scale center.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Source label e.drift.general: diagonal entries of the terminal-normalized
annealed drift Ahom_m^{-1/2} (Ahom_j - Ahom_m) Ahom_m^{-1/2}.
Equations
- HCPolySupport.HighContrast.EntryScale.terminalAnnealedFullBlockDriftDiagAtScales hP hStruct j m (Sum.inl val) = hP.barSigmaAtScale hStruct ↑j / hP.barSigmaAtScale hStruct ↑m - 1
- HCPolySupport.HighContrast.EntryScale.terminalAnnealedFullBlockDriftDiagAtScales hP hStruct j m (Sum.inr val) = (hP.barSigmaStarAtScale hStruct ↑j)⁻¹ / (hP.barSigmaStarAtScale hStruct ↑m)⁻¹ - 1
Instances For
Source label e.drift.general: the deterministic annealed-drift matrix is
exactly diagonal after terminal scalar normalization.
Source label e.drift.general: deterministic annealed-drift norm
D_{j,m} from the source proof.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Source label e.drift.general: D_{j,m} is nonnegative.
Source label e.drift.general: bound D_{j,m} by uniform bounds on the two
scalar diagonal entries.
Source label e.drift.general: source-facing scalar-ratio form of the
pointwise deterministic drift bound.
Source label e.drift.general: the pointwise deterministic drift is bounded
by the product excess of the two scalar terminal ratios.
Source label e.drift.general: under (P4) and j <= m, the scalar-chain
monotonicity supplies the hypotheses for the product-excess drift bound.
Source label e.drift.general: library-facing pointwise drift bound in the
paper's scalar excess notation F_n = Theta_n - 1.
Source label e.tau.sum.absorb: on a no-drop window [k,m], every
intermediate contrast drop is bounded by the endpoint no-drop budget.
Source label e.drift.general: on a no-drop window, the pointwise drift
bound becomes the paper's no-drop form with F_m = Theta_m - 1.
Source label l.union.bound: the concrete terminal/intermediate diagonal
normalization-change matrix has full-block operator norm at most
sqrt widetildeTheta_0.
Source label l.union.bound: changing both sides of a full-block observable
from intermediate scale j to terminal scale m costs at most the corrected
initial contrast budget T = widetildeTheta_0.
Source label l.union.bound: terminal normalization of a block centered at
scale j costs at most T = widetildeTheta_0 times the same centered block
with its intermediate-scale normalization.