Reusable proof technology
Shared identities, explicit consumers and separate adapters. These targets do not reschedule the active mathematical frontier.
RED · reusable target, not a compiled leaf
Density smoothing and score regularity
Normalized convolution, conditional score/covariance representatives and controlled higher derivatives.
Source: 2609.06906v1. Section 3.1; Lemmas 4.1–4.3; Appendix C
Parents: Independent root candidate
Consumers: Picard trajectory and quadrature error
Failure boundary
Not potential averaging; not an exactly unbiased score oracle.
Existing Lean reuse-search locations
- AutoSamplingTheory.TechnicalLemmas.Measure.GaussianSmoothing — search candidate only
Inspect exact ASTIS and pinned Mathlib types before declaring reuse. No new source theorem or wrapper is introduced by this list.
RED · reusable target, not a compiled leaf
Picard trajectory and quadrature error
Integral Hamiltonian flow, local bias/variance and Wp propagation with finite moments.
Source: 2609.06906v1. Sections 3.2, 4–5; Appendix B
Parents: Density smoothing and score regularity
Consumers: Truncation and reverse transport
Failure boundary
Higher derivatives of the smoothed density do not strengthen assumptions on the original potential for free; oracle bias must be paid.
Existing Lean reuse-search locations
- AutoSamplingTheory.TechnicalLemmas.Analysis.Calculus.Taylor — search candidate only
Inspect exact ASTIS and pinned Mathlib types before declaring reuse. No new source theorem or wrapper is introduced by this list.
RED · reusable target, not a compiled leaf
Truncation and reverse transport
A probability-law witness with separate TV distance and Renyi certificate, using p-moment and heat-flow inputs.
Source: 2609.06906v1. Lemmas 6.2–6.3; Theorem 7.1(ii)
Parents: Picard trajectory and quadrature error · Quadratic RGO composition
Consumers: Proxy-stable kernel composition
Failure boundary
An actual sample law is not its comparison witness; smoothing and recursion cannot be removed.
Existing Lean reuse-search locations
- AutoSamplingTheory.TechnicalLemmas.InformationTheory.Renyi — search candidate only
- AutoSamplingTheory.TechnicalLemmas.Measure.GaussianSmoothing — search candidate only
Inspect exact ASTIS and pinned Mathlib types before declaring reuse. No new source theorem or wrapper is introduced by this list.
RED · reusable target, not a compiled leaf
Quadratic RGO composition
Complete the square once; keep potential equality up to a constant separate from equality of normalized conditional kernels.
Source: 2609.06906v1 · 2609.06905v1. SPHMC Lemma 6.4; PBPS Section 2.2
Parents: Independent root candidate
Consumers: Truncation and reverse transport · Conditional harmonic flight and bounce kernel
Failure boundary
Positive variance, normalization and parameter measurability are real obligations.
Existing Lean reuse-search locations
- AutoSamplingTheory.TechnicalLemmas.Probability.ConditionalKernel — search candidate only
Inspect exact ASTIS and pinned Mathlib types before declaring reuse. No new source theorem or wrapper is introduced by this list.
RED · reusable target, not a compiled leaf
Two reflections with different roles
Velocity reflection R_h with R_0=I; separately the augmented involution (x,y) ↦ (x,2x-y).
Source: 2609.06905v1. (2.4); Proposition 2.1; Section 3.2
Parents: Independent root candidate
Consumers: Conditional harmonic flight and bounce kernel · Micro–macro modified-L2 contraction
Failure boundary
Isometry or involution alone does not prove process invariance, ergodicity or convergence.
Existing Lean reuse-search locations
- AutoSamplingTheory.TechnicalLemmas.Geometry.EuclideanSpaceCoordinates — search candidate only
Inspect exact ASTIS and pinned Mathlib types before declaring reuse. No new source theorem or wrapper is introduced by this list.
RED · reusable target, not a compiled leaf
Conditional harmonic flight and bounce kernel
Non-explosive event process, path reversal, conditional invariance and the averaged half-turn kernel.
Source: 2609.06905v1. Propositions 3.1–3.2; Appendix A
Parents: Two reflections with different roles · Quadratic RGO composition
Consumers: Micro–macro modified-L2 contraction
Failure boundary
Not a free-flight vanilla BPS semigroup; preserve fixed-reference and zero-rate conventions.
Existing Lean reuse-search locations
- AutoSamplingTheory.TechnicalLemmas.Probability.ConditionalKernel — search candidate only
Inspect exact ASTIS and pinned Mathlib types before declaring reuse. No new source theorem or wrapper is introduced by this list.
RED · reusable target, not a compiled leaf
Micro–macro modified-L2 contraction
Conditional-expectation projection, reflection coupling, operator interpolation and norm-equivalent modified energy.
Source: 2609.06905v1. Theorem 3.5; Appendices B–D
Parents: Two reflections with different roles · Conditional harmonic flight and bounce kernel
Consumers: Implementation coupling and query caps
Failure boundary
Do not import reversible Dirichlet coercivity or continuous-time hypocoercivity without an explicit discrete adapter.
Existing Lean reuse-search locations
- AutoSamplingTheory.TechnicalLemmas.FunctionalInequalities.SemigroupDecay — search candidate only
Inspect exact ASTIS and pinned Mathlib types before declaring reuse. No new source theorem or wrapper is introduced by this list.
RED · reusable target, not a compiled leaf
Implementation coupling and query caps
Ideal mixing, RGO error, solver fallback, event-rate cap and actual-input expected query bounds.
Source: 2609.06905v1. Theorem 4.3; (4.17)–(4.18)
Parents: Micro–macro modified-L2 contraction
Consumers: Proxy-stable kernel composition
Failure boundary
Exact ideal invariance and approximate implemented accuracy remain distinct.
Existing Lean reuse-search locations
- AutoSamplingTheory.TechnicalLemmas.Probability.ConditionalKernel — search candidate only
Inspect exact ASTIS and pinned Mathlib types before declaring reuse. No new source theorem or wrapper is introduced by this list.
RED · reusable target, not a compiled leaf
Proxy-stable kernel composition
TV data processing plus triangle inequality; separate conditional expected-cost addition.
Source: 2609.06906v1 · 2609.06905v1. SPHMC Section 7.2; PBPS (4.18)
Parents: Truncation and reverse transport · Implementation coupling and query caps
Consumers: Composition view
Failure boundary
Markov-kernel measurability and mass one; no unbounded-cost transfer from TV alone.
Existing Lean reuse-search locations
- AutoSamplingTheory.TechnicalLemmas.Probability.ConditionalKernel — search candidate only
Inspect exact ASTIS and pinned Mathlib types before declaring reuse. No new source theorem or wrapper is introduced by this list.
Next dependency-ready packet candidates
Branch persistence and finite-depth RGO parameter threshold
RGOCalculus and both scalar regime estimates now compile; exact independent admission is tracked in their cells. RecursiveVariance includes zero initial precision and a guarded finite-previous-variance comparison, with c<1/4 generalized away only for the selected scalar formula. Next inspect the actual condition-number update for branch persistence, then geometric progress toward the finite positive threshold with exact stage indexing. These successor claims remain planned, not proved.
Exclude: Do not relax the algorithm's source schedule or infer termination directly from a single-step inequality. Recursive error, terminal FORS work, reference-point cost, implemented sampling and actual-input expected query costs remain independent. TV proximity does not transfer unbounded expected cost.
Source-detail audits of analytic and process consumers
Separate conditional score/covariance representatives, domination, event-process domains and discrete modified-energy contracts. Reuse the existing Gaussian law, reflection and normalization results.
Exclude: Preserve earlier Chewi 8.4.1 evidence; do not require the whole SDE library before reachable paper lemmas or infer process invariance from reflection algebra.