Section 12 pasting: degree-zero branch #
Auxiliary constructions for the d = 0 complementary branch of thm:ld-pasting: the
counterpart of MIPRE/Background/LIDT/MIPStarRE/LDT/Pasting/Bernoulli/DegreeZero.lean in the port
of planning/c6b-plan.md (milestone M11, section "Port conventions").
A strategy is a SymStrat params.next 𝔓 K and a slice family an IdxPolyFamily params 𝔓; the
candidate is the averaged slice family viewed as a polynomial in one more variable,
averagedSliceAppendedSubMeas, and the consistency statements are ConsRels on
strategy.state. Three uses of the vendored strategy fields become theorems of the model:
- the bounds of the two failure probabilities by
1(vendored throughstrategy.isNormalized) areS.bipartiteConsError_uniform_le_oneandPasting.bipartiteSSCError_uniform_le_one; - the swap of the consistency relation (vendored through
strategy.densityFixed) isS.consRel_symm_of_density_fixed, and the triangle stepPreliminaries.triangleSub_righttakes no normalization argument; - the residual mass
ev(R(1 − H.total)) = 1 − ev(L H.total)(vendored throughstrategy.permInvState.swap_ev, an entrywise Kronecker identity andev_one_of_isNormalized) isS.ev_L_eq_ev_R,S.leftTensor_subandS.ev_one_of_isNormalized. No statement carries a swap or normalization hypothesis, the vendored statements having none.
The proofs are shorter than the vendored ones. The good-strategy forms
degreeZero_averagedSlice_liftedVerticalLineConsistency,
degreeZero_averagedSlice_pointConsistency and degreeZeroPastedPointConsistency are their
axis/self-consistency forms applied to hgood.axisParallelTest and hgood.selfConsistencyTest,
so each is declared after its _of_axis_self form, the reverse of the vendored order. The
height averaging of the lifted-line estimate decomposes both sides through
avgOver_uniform_pointNext_decompose once each.
Not ported #
References #
In LionSR/MIPStarRE at commit 507e8122, not in this repository:
references/ldt-paper/ld-pasting.texblueprint/src/chapter/ch09_pasting.tex
Upstream keeps this helper private since its Lean-module port (LionSR/MIPStarRE at 5fc363b); the port carries its own copy, with upstream's proof.
Scalar absorption for the degree-zero submeasurement consistency error.
Evaluating the degree-zero appended-slice candidate is the height average of the original evaluated slice family at the same old point.
The point-consistency hypothesis may be truncated at the trivial unit bound.
Axis/self-consistency form of the degree-zero lifted-line consistency estimate.
The degree-zero averaging argument uses the axis-parallel test and self-consistency, but not the diagonal-line test.
The averaged degree-zero pasted submeasurement is consistent with the lifted vertical-line answers.
Axis/self-consistency form of the degree-zero point-consistency estimate before completion.
The averaged degree-zero pasted submeasurement is point-consistent before completion.
Axis/self-consistency form of the degree-zero point-consistency construction.
This is the same construction as degreeZeroPastedPointConsistency, with the
ordinary good-strategy hypotheses replaced by the two estimates actually used in
the proof.
Degree-zero point-consistency construction for thm:ld-pasting.
Paper origin: references/ldt-paper/ld-pasting.tex:12-55. In the degree-zero
branch the slice polynomials and the last-coordinate line answers are constant
on their respective domains. The measurement is the completion of
averagedSliceAppendedSubMeas, the averaged slice family viewed as a global
polynomial family by ignoring the appended variable.