Skip to content

Latest commit

 

History

History
151 lines (138 loc) · 16.7 KB

File metadata and controls

151 lines (138 loc) · 16.7 KB

Term Map

This is the concise concept-to-module routing index, including public entry points. The old detailed map was collapsed into archive.md; use git history for exact old wording. Exact RandomMatrix endpoints, imports, and assumptions are in RandomMatrixAPI.md; supported scope is maintained in Status.md, active work in TODO.md, ownership and dependency direction in RandomMatrixArchitecture.md, and family-level proved/open status in TheoremAtlas.md.

Area Main terms Source
Probability basics Event, ProbabilityMeasure, RandomVariable, law, expect HighDimProb
Finite expectation sums expect_finset_sum HighDimProb/Expectation.lean
Tail vocabulary upperTailEvent, lowerTailEvent, absTailEvent, tail probabilities HighDimProb/Tail.lean
Scalar size realLpNorm, HasFiniteMoment, SubGaussianTail, SubExponentialTail, Psi2Bound, Psi1Bound HighDimProb/Scalar and concentration files
Scalar concentration Markov, Chebyshev, Orlicz/tail, moment, MGF, Rademacher, Hoeffding, Bernstein routes HighDimProb/Concentration
Hanson-Wright centeredQuadraticForm, deterministicFrobeniusNorm, HighDimProb.HansonWright.hanson_wright_inequality_hdp; finite real matrix and finite coordinate family, K > 0, iIndepFun, coordinate HasSubgaussianMGF at K^2, universal-constant Frobenius/operator min-form tail, no symmetry premise HighDimProb/Concentration/HansonWright.lean · HighDimProbTest/HansonWrightAPI.lean
Analysis helpers real inequalities including exp_mul_le_chord_exp_of_nonneg_of_le HighDimProb/Analysis/RealInequalities.lean
Analysis dense-sup bridge ciSup_eq_ciSup_of_denseRange HighDimProb/Analysis/DenseSup.lean
Sum/integral comparison sum_mul_sub_le_intervalIntegral_of_antitoneOn, tendsto_intervalIntegral_of_leftEndpoint_tendsto, le_intervalIntegral_of_le_residual_add_of_tendsto_zero HighDimProb/Analysis/SumIntegral.lean
Compact residual convergence tendstoUniformlyOn_abs_sub_of_isCompact, tendsto_edist_uniformFun_abs_sub_of_isCompact, tendsto_toReal_edist_uniformFun_abs_sub_of_isCompact HighDimProb/Analysis/CompactApproximation.lean
Dominated expectation convergence MeasureTheory.tendsto_integral_filter_of_dominated_convergence (direct Mathlib use; no project wrapper) HighDimProbTest/ExpectationConvergenceAPI.lean
Random families/processes RandomFamily, RealRandomFamily, IsRandomFamily, familyAt, mapRandomFamily, RandomProcess, IsRandomProcess, processAt, RandomSample, IsRandomSample, sampleEvaluation HighDimProb/Process.lean
SubGaussian process increments HasSubGaussianMGFIncrements, hasSubgaussianMGF_mono, HasSubGaussianMGFIncrements.centeredSubGaussianMGF_of_dist_le HighDimProb/SubGaussianProcess.lean and HighDimProb/SubGaussian.lean
Random vectors random-vector, covariance, isotropic, subGaussian-vector vocabulary HighDimProb/Vector.lean
Geometry nets, metric entropy, Gaussian width vocabulary HighDimProb/Geometry.lean
Finite process supremum processSup, isRandomVariable_processSup, integrable_processSup HighDimProb/Chaining.lean
Cover and packing vocabulary epsilonRadius, IsEpsilonNet, IsInternalEpsilonNet, IsEpsilonSeparated, externalCoveringNumber, coveringNumber, packingNumber HighDimProb/Nets.lean and HighDimProb/MetricEntropy.lean
Finite chaining chain_sub_eq_sum_range, norm_sub_chain_le_sum_of_step_bound, norm_sub_chain_le_sum_of_level_sup, expect_abs_sub_chain_le_sum_of_level_sup, finset_sup'_norm_sub_anchor_le_residual_add_sum_of_step_bound, expect_finset_sup'_abs_sub_anchor_le_residual_add_sum_of_step_bound HighDimProb/Chaining.lean
Deterministic finite LogSumExp sum_exp_pos, exp_mul_sup'_le_sum_exp, sup'_le_log_sum_exp_div, log_sum_exp_le_log_card_add HighDimProb/Analysis/LogSumExp.lean
Fixed-CGF finite maximum expect_processSup_le_of_cgf_bound_at HighDimProb/Concentration/FiniteMax.lean
Optimized subGaussian maxima CenteredSubGaussianMGF.neg, expect_processSup_le_of_centeredSubGaussianMGF, expect_finset_sup'_abs_le_of_centeredSubGaussianMGF HighDimProb/SubGaussian.lean and HighDimProb/Concentration/SubGaussianMax.lean
Finite subGaussian chaining finiteEntropySum, expect_abs_sub_chain_le_sum_of_level_sup_of_centeredSubGaussianMGF, expect_abs_sub_chain_le_sum_of_level_sup_of_centeredSubGaussianMGF_of_card_le, expect_abs_sub_chain_le_sum_of_level_sup_of_subGaussianMGFIncrements, expect_abs_sub_chain_le_finiteEntropySum, expect_abs_sub_chain_le_finiteEntropySum_of_path, expect_finset_sup'_abs_sub_anchor_le_finiteEntropySum, expect_finset_sup'_abs_sub_anchor_le_truncatedEntropyIntegral, expect_abs_sub_dyadic_path_le_truncatedEntropyIntegral, finiteEntropySum_dyadic_le_four_mul_intervalIntegral_coveringNumber HighDimProb/MetricEntropy.lean and HighDimProb/Concentration/SubGaussianMax.lean
Dense/full anchored supremum passage expect_iSup_abs_sub_anchor_le_of_denseRange_of_prefix_bound, expect_iSup_abs_sub_anchor_le_mul_intervalIntegral_of_denseRange_of_prefix_bound (the latter requires supplied prefix bounds and supplied residual expectation convergence) HighDimProb/Concentration/SubGaussianMax.lean
Dudley entropy integral dudleyEntropyIntegral (anchored supremum over a totally bounded subtype, under explicit dense-sequence, anchor-net, regularity, and integrability assumptions) HighDimProb/Concentration/Dudley.lean
Volumetric covering bounds l1Ball, coveringNumber_euclideanBall_le, coveringNumber_l1Ball_le (internal ENat bounds ceil((1 + 2R/eps)^card) and ceil((1 + 4R/eps)^card), for R >= 0, eps > 0, and finite nonempty index; the l1 bound is via B1 ⊆ B2 (l1Ball inside the Euclidean closed ball) plus Mathlib subset comparison, not the sharper Maurey estimate) HighDimProb/Geometry/CoveringNumber.lean; public import HighDimProb.Geometry
ENat covering-number bridge coveringNumber_le_card_of_isInternalEpsilonNet, exists_nat_eq_coveringNumber_of_isInternalEpsilonNet, exists_finset_isInternalEpsilonNet_of_totallyBounded, exists_parentMap_of_subset_of_isInternalEpsilonNet, exists_finset_parentMap_of_internalLevels, exists_finset_parentMap_of_internalRadiusLevels, exists_finset_path_of_parentMap, exists_finset_internalNetFamily_parentMap_path_of_totallyBounded, dyadicRadius, tendsto_dyadicRadius_atTop, tendsto_intervalIntegral_dyadicRadius_atTop HighDimProb/MetricEntropy.lean
Random matrices random matrix families, self-adjointness, sums, operator norm, spectral events, ordered spectral endpoints (lambdaMaxOrdered, lambdaMinOrdered) HighDimProb/RandomMatrix
Matrix concentration public trace-MGF, tail, Matrix Bernstein, and sample-covariance facade HighDimProb.RandomMatrix.Concentration
Matrix sub-Gaussian route MatrixSubGaussianMGF, MatrixSubGaussianMGF.neg, traceMGFVarianceProxyBound_of_matrixSubGaussian_under_troppPrimitive, subGaussian_quadraticFormUpperTail_under_troppPrimitive HighDimProb.RandomMatrix.Concentration, SubGaussian.lean, SubGaussianMatrixAPI.lean
Directional matrix sub-Gaussian route DirectionallySubGaussianSelfAdjointMatrix, IsUnitSphereNet, independent-sum closure, deterministic net control, and explicit-N.card operator-norm tails HighDimProb.RandomMatrix.Concentration, DirectionalSubGaussian.lean, DirectionalOperatorNorm.lean
Matrix analysis providers matrix exponential/logarithm calculus, resolvents, relative entropy, Lieb/Epstein, Golden--Thompson Provider.Analysis
Matrix conditioning providers kernels, frozen-parameter conditional expectation, natural histories Provider.Conditioning
Matrix concentration providers (internal/expert) integrability compression, trace-MGF, tails, scoped Matrix Bernstein Provider.Concentration
Matrix zero variance zero variance proxy, almost-everywhere zero summands and sums, null positive operator-norm tails HighDimProb/RandomMatrix/VarianceZero.lean
Matrix Bernstein trace-MGF/Tropp bundles, scalar threshold inversion, MatrixBernstein.*_of_primitives optimized/operator-norm/high-probability facades, Bernstein CFC hardbone, variance-proxy bridges, centered-square exact-row adapters, support/effective-rank trace bridges, prefix/reindex/negative adapters, and compact sample-covariance contracts RandomMatrixAPI.md
PrecisionDA applications deterministic column-sample covariance, leave-one-out covariance, shrinkage resolvents, rank-one/Woodbury identities, Frobenius trace-expansion wrappers, and H1/H2/Theorem 1 provider-contract vocabulary HighDimProb/Applications/PrecisionDA
Examples compact statement-route index plus representative sample covariance, random-feature, gradient, NTK, LoRA, attention, Fisher, natural-Tropp, and PrecisionDA routes HighDimProb/Examples

The matrix sub-Gaussian predicate is only a conditional Loewner MGF contract. The finite-family and tail declarations retain the Tropp comparison, random/self-adjoint/independence, variance-proxy self-adjointness, and unbounded matrix/trace-exponential integrability assumptions; the tail endpoint additionally retains the aggregate proxy spectral bound at theta = t / sigmaSq. They do not bundle centeredness, PSD, or probability, and do not assert unconditional Matrix Chernoff, full Tropp, or infinite-dimensional results.

Lookup Rule

Use this file for orientation only. For exact declarations, use doc-gen output, #check, or source search. Keep new entries short and link to the source rather than copying full theorem signatures.

Public And Expert RandomMatrix Terms

Use HighDimProb.RandomMatrix.Concentration for downstream matrix-concentration results. The HighDimProb.RandomMatrix.Provider.Analysis, .Conditioning, and .Concentration imports are internal/expert proof boundaries; use the narrowest provider layer needed according to the dependency ownership documented in RandomMatrixArchitecture.md. The broad expert HighDimProb.RandomMatrix.Provider facade imports all three. HighDimProb.RandomMatrix.LiebProvider remains a compatibility import and should not own new declarations.

The analysis layer contains ambient and self-adjoint carrier matrix-exp Frechet derivative primitives, the scalar divided-difference coefficient matrixExpDividedDifferenceSeries, the preferred trace-pairing alias MatrixExpFDeriv.conjDiagonalSymmTraceSum, the strictly-positive carrier CFC.log first-derivative namespace CFCLog, preferred diagonal adapters CFCLog.diagonalDerivEntryMul, CFCLog.diagonalLineDerivEntryMul, and CFCLog.diagonalLineDerivTraceSum, the short inverse/trace-resolvent derivative layer, the finite-cutoff log-resolvent namespace LogResolvent, and the inverse-convexity quadratic-form and segment identities inv_quadraticForm_affine_le_of_posDef, inv_quadraticForm_iSup_affine_of_posDef, and inv_matrixLE_convex_combo_le_of_posDef, plus the relative-entropy route APIs RelativeEntropy.leftRightDenominatorMatrix, RelativeEntropy.leftRightRelativeEntropyIntegrand, RelativeEntropy.leftRightRelativeEntropyIntegrand_jointConvex, RelativeEntropy.traceMatrixRelativeEntropyPlain, LeftRightRelativeEntropyIntegrandDensityIntegrable, TraceMatrixRelativeEntropyPlainLeftRightIntegralRepresentation, leftRightRelativeEntropyIntegrandDensityIntegrable, traceMatrixRelativeEntropyPlainLeftRightIntegralRepresentation, relativeEntropyJointConvexity_of_leftRight_density_integral_representation, relativeEntropyJointConvexity_of_leftRight, liebTraceExpConcavity_statement_of_leftRight_density_integral_representation, liebTraceExpConcavity_statement_of_leftRight, epsteinAffineLineConcavity_of_leftRight_density_integral_representation, epsteinAffineLineConcavity_of_leftRight, RelativeEntropy.scalarTerm_nonneg, RelativeEntropy.diagonalTerm_nonneg, RelativeEntropy.fullMatrixKlein_nonneg_of_isHermitian_of_strictlyPositive, RelativeEntropy.logShift, RelativeEntropy.expLogMatrix, gibbsVariationalUpperBoundPremise_of_fullMatrixKlein, and RelativeEntropy.fullKlein_epsteinConcavity.

Provider.Analysis also exposes the Epstein consumer namespace EpsteinLine, left/right Epstein/Lieb and relative-entropy/Gibbs bridges, spectral endpoint monotonicity, trace-exp domain positivity, CFC-log resolvent cutoff/remainder bridges, and fixed-numerator trace-resolvent convexity.

Provider.Conditioning owns conditioning-kernel reductions and the TroppNaturalHistory.* aliases for suffix measurability and strengthened history/current-step independence. Long natural-history theorem names remain compatibility surfaces.

Provider.Concentration owns the Tropp one-step and trace-MGF-to-Laplace contracts, provider-compressed natural-state tail helpers, generated-history Bernstein finite-family/trace-MGF/tail assembly, the restricted-history TraceExpConditioning.bernsteinInputs_of_primitives and TraceExpConditioning.bernsteinStep_of_history_le facades, integrability compression, support-to-excess compression, and tail-event subset discharge. Downstream callers consume re-exported results through HighDimProb.RandomMatrix.Concentration.

Matrix Bernstein provider-compressed tail names include MatrixBernsteinConditioningTraceMGFProviderAssumptions, matrixBernsteinQuadraticFormUpperTail_of_conditioningTraceMGFProviderAssumptions, matrixBernsteinQuadraticFormUpperTail_of_naturalStateProviderAssumptions, and their *_tailSubsetDischarged_of_randomSelfAdjoint TailEvent wrappers.

The main-layer MatrixBernstein.optimized_of_primitives and MatrixBernstein.highProbability_of_primitives facades construct the finite-family generated-history witness and consume the scalar inversion layer. MatrixBernstein.centeredRankOneExactRow and MatrixBernstein.sampleCovarianceExactRow additionally close the row-specific variance and normalized sample-covariance tail composition. MatrixBernstein.centeredRankOneExactRowHighProbability supplies the canonical high-probability threshold for the unnormalized centered rank-one sum. MatrixBernstein.sampleCovarianceExactRowHighProbability evaluates that route at the canonical threshold divided by the row count, while iIndepFun_centeredRankOne transfers raw vector-family independence to the centered outer-product family. They do not prove unconditional Matrix Bernstein; exact caller assumptions are recorded in RandomMatrixAPI.md, and family-level closure is recorded in TheoremAtlas.md.