The six-term skyscraper fragment (CC8, D7, proof plan §6.9(c)-(g)) #
Unit: cech-cohomology (docs/design/cech-cohomology.md §4.7, §6.9).
C1.retype_mem_Z1': the general (non-coboundary) retype-preserves-Z1fact.h1CoverIncl_mk_retype: theD-inclusion of a retypedZ1 D 𝒰class recovers the originalZ1 D' 𝒰class.memLD_of_isAdapted: on a cover adapted todiffSupp D D', everyZ1 D' 𝒰-cocycle already satisfies the (smaller)D-bound — diagonal components vanish to order⊤(Z1.ord_diag), off-diagonal components avoid the finite set whereD ≠ D'(FinCover.IsAdapted.not_mem_inf), soD = D'pointwise there.H1Incl_surjective: noH², no long exact sequence, no snake lemma — every class ofH1 D'is already represented, on a suitably adapted cover, by a genuineZ1 D-cocycle.exists_tail_approx("Lemma B"): any germ nearqof order≥ -d'is approximated, to order> -d - 1atq, by a chart-source germ inordGe q (-d')— a finite Laurent tail, built by iterating the one-step leading-coefficient subtraction (exists_tail_step; same correction step as WindowRank's splitting, redone on an arbitrary openW ∋ qbecauseleadCoeffis chart-source-bound).windowDefect+Realizes(§6.9(c), D7): the pointwise-ordrealization predicate. The design's(w q).outform is replaced by quantification over all representatives ofw q(equivalent bywindowDefect_bound_of_mk_eq; strictly easier to consume).exists_realization(§6.9(c)) andmlClass_eq_of_realizes(§6.9(d), "Lemma A" — proved without any adaptedness hypothesis: only theRealizesbounds andD = D'offdiffSuppenter).windowConnect/windowConnect_spec(§6.9(d)): the connecting mapWindow D D' →ₗ H¹(D).exact_windowMap_windowConnect(§6.9(e)),exact_windowConnect_H1Incl(§6.9(f)): withWindow.lean'sexact_inclusion_windowMap/inclusion_injectiveandH1Incl_surjectivebelow, this completes the six-term fragment0 → L(D) → L(D') → Window D D' → H¹(D) → H¹(D') → 0.
General (non-coboundary) version of C1.retype_mem_Z1: retyping a Z1 D'-cocycle whose
components all satisfy the D-bound gives a Z1 D-cocycle.
The D-inclusion of a retyped Z1 D 𝒰 class recovers the original Z1 D' 𝒰 class.
Small order arithmetic helpers #
Local tail approximation ("Lemma B", §6.9(c) input) #
A germ γ near q of order ≥ -d' at q is approximated to order ≥ -d by a genuine
chart-source germ in ordGe q (-d') — a finite Laurent tail. Built by iterating the one-step
leading-coefficient subtraction (the same correction as WindowRank.lean's splitting, redone
here on an arbitrary open W ∋ q because leadCoeff is chart-source-bound).
Local tail approximation ("Lemma B"): a germ γ on an open W ∋ q inside the chart
source with ord_q γ ≥ -d' is matched, to order ≥ -d at q, by a chart-source germ of
ordGe q (-d') (a finite Laurent tail Σ c_m θ_{q,m}).
The window defect (D7): the pointwise-ord comparison germ #
The comparison germ of a local section γ (on a member V) against a chart-source germ
ψ at q, on the common open V ∩ (chartAt ℂ q).source (D7: all realization bookkeeping is a
pointwise ord bound on this germ).
Equations
- RS.Cech.windowDefect γ ψ = (RS.MeroGermOn.restrict ⋯) γ - (RS.MeroGermOn.restrict ⋯) ψ
Instances For
The ord of the defect at q can be computed after restricting both germs to any open
T ∋ q inside both domains (the normalization workhorse for all defect bookkeeping).
Restricting the local section to a smaller member does not change the defect's ord.
Rep-change: the defect bound only depends on the WindowAt-class of the chart-source germ
(two representatives differ by ordGe q (-dq), which is absorbed by ord_add). This is why
Realizes may quantify over all representatives.
C1.MemLD closure properties #
On a cover adapted to diffSupp D D', every Z1 D'-cocycle already satisfies the D-bound
componentwise: diagonal components vanish to order ⊤, off-diagonal components avoid the finite
set where D ≠ D' (adaptedness), hence D = D' there and the D'-bound is the D-bound.
g realizes the window vector w (pointwise-ord form, D7): at every q of
diffSupp D D' and every member containing q, the defect of g's component against any
representative of w q has order ≥ -(D q) at q. (Deviation from the design's (w q).out
formulation: quantifying over all representatives is equivalent by
windowDefect_bound_of_mk_eq, and strictly easier to consume.)
Equations
- One or more equations did not get rendered due to their size.
Instances For
Lemma A (§6.9(d)): independence of the realization #
Lemma A (§6.9(d)): any two Mittag-Leffler realizations of the same window vector give
the same H¹(D)-class. No adaptedness is needed: at diffSupp-points the two Realizes
bounds control the difference, everywhere else D = D' and the D'-bounds do.
exists_realization (§6.9(c)) #
Adapted-cover realization (§6.9(c)): every window vector is realized by a
Mittag-Leffler 0-cochain on a cover adapted to diffSupp D D' — on the (unique) member
through q, the restriction of a chart-source representative of w q, shrunk into a pole-free
zone avoiding all other points of diffSupp D D' ∪ supp D ∪ supp D'; 0 elsewhere.
The connecting map windowConnect (§6.9(d)) #
The first-cohomology class associated to a section in the divisor window.
Equations
- RS.Cech.windowConnectRaw h w = RS.Cech.mlClass ⋯.choose ⋯.choose ⋯
Instances For
The connecting map δ of the six-term skyscraper fragment (§6.9(d)): realize the
window vector on an adapted cover (exists_realization), take the Mittag-Leffler class
(mlClass); well-defined by Lemma A (mlClass_eq_of_realizes), which also gives linearity.
Equations
- RS.Cech.windowConnect h = { toFun := RS.Cech.windowConnectRaw h, map_add' := ⋯, map_smul' := ⋯ }
Instances For
windowConnect agrees with the Mittag-Leffler class of any realization (the working
form of the connecting map — this is what consumers should use).
Exactness (§6.9(e)/(f)) #
Exactness at Window D D' (§6.9(e)): windowConnect h w = 0 iff w is the window
vector of a global section of L(D').
Exactness at H¹(D) (§6.9(f)): the kernel of H1Incl is exactly the image of the
connecting map — toH1_injective (Forster 12.4) turns the vanishing into a D'-coboundary
witness on the representing cover itself, and exists_tail_approx reads off its window
vector.