Preprint
Article

This version is not peer-reviewed.

Singular Hodge Atoms at Conifold Degenerations: Recollement, Rees Non-Splitting, Exceptional-Curve Relations, and Hecke Costalk Detection

Submitted:

16 November 2026

Posted:

18 November 2026

Read the latest preprint version here

Abstract
We define a marked singular conifold Hodge atom as the extension-bearing mixed-Hodge-module datum comprising the intersection-complex constituent, the (-2)-twisted punctual quotient, and their specified attaching morphism. The construction uses open--closed recollement rather than spectral decomposition; it is not an identification with a Katzarkov–Kontsevich–Pantev–Yu spectral atom. For the determinantal ordinary double point, a filtered localization model proves that the canonical and every nonzero corrected local attachment remain nonsplit after associated-graded Rees specialization. When a projective small resolution with smooth source exists, codimension-two purity and the Gysin map identify the rational global attaching boundary with the signed exceptional-curve class map, up to the marked diagonal rescaling. For successive framed rank-two Hecke modifications, we compute the complete scheme-theoretic punctual first-jet incidence, its reduced mixed Hodge cohomology and terminal-lattice images, and establish naturality of the transported extension under the specified two-step composition equivalence. An extraordinary-costalk theorem proves nonvanishing under explicit properness, smooth-purity, node-factorization, and constant-kernel hypotheses; their verification on the genuine Higgs charts and the numerical perverse boundary maps remain open. A synthesis theorem collects these separately scoped results. Finally, an exact filtered-flat Rees functor produces formal F-bundles without asserting an unconstructed singular-to-quantum comparison.
Keywords: 
;  ;  ;  ;  ;  ;  ;  ;  ;  ;  ;  ;  ;  

MSC:  14F43; 14D24; 14J32; 18G80; 22E57; 32S40

1. Introduction

A degeneration of smooth projective threefolds to a nodal central fiber has a local perverse contribution absent at a smooth point. Its defining information is not exhausted by nearby- and vanishing-cycle dimensions: the corrected mixed-Hodge-module construction retains an attaching morphism from a punctual quotient to the shifted intersection-complex object. We call the resulting marked extension-bearing datum a singular conifold Hodge atom, as precisely defined in Definition 15. This is a non-spectral construction internal to mixed Hodge modules: its constituents, gluing arrow, endpoint markings, and cone normalization are geometric input, not spectral factors selected by an Euler operator. The purpose of this article is to compute properties of this selected datum under open–closed recollement, filtered differential-module realization, and framed Hecke correspondences, while retaining the actual morphism rather than only its class in a Grothendieck group. The main results are a conifold Rees non-splitting theorem, a global exceptional-curve boundary identification under a projective small-resolution hypothesis, a complete scheme-theoretic two-step Higgs first-jet incidence calculation, and an extraordinary-costalk nonvanishing theorem with explicit additional geometric hypotheses. Theorem 7 collects the resulting properties without merging their distinct domains or coefficient categories.
A separate comparison problem concerns the spectral Hodge atoms of Katzarkov–Kontsevich–Pantev–Yu (KKPY) [1]. A singular conifold Hodge atom is a marked extension datum in D b MHM ( X 0 ) , whereas a KKPY atomic representative is a meromorphic F-bundle on a specified non-archimedean parameter base satisfying additional spectral and equivariance conditions. Neither the shared word “atom” nor a common Hodge-theoretic invariant identifies these constructions. We establish an exact Rees functor from Griffiths-transverse filtered flat bundles to formal F-bundles and calculate its leading operators, but that functor has a different input type from the singular filtered D -module. The desired extension-sensitive singular-to-quantum comparison remains an independently typed construction problem; it is not used to prove the conifold results.

1.1. The Specified Conifold Attachment

Let π : X → Δ be the analytic restriction of the proper algebraic family fixed in Section 2.1. Its general fibers are smooth threefolds and its reduced central algebraic fiber X 0 has a finite ordinary-double-point locus Σ . Put U = X 0 ∖ Σ , with j : U ↪ X 0 and i p : { p } ↪ X 0 . The corrected construction of [2], under its stated hypotheses and conventions, supplies the short exact sequence in MHM ( X 0 )
0 ⟶ I H → u P H → v R Σ H ⟶ 0 ,
where
I H = IC X 0 H , R Σ H = ⨁ p ∈ Σ i p * Q H ( − 2 ) .
Its fixed Yoneda connecting morphism is
ϵ Σ : R Σ H ⟶ I H [ 1 ] in D b MHM ( X 0 ) .
The marked singular conifold Hodge atom is therefore A Σ H = ( I H , R Σ H , ϵ Σ ; u , v ) , with the maps of (1) included in its marking. Its persistent intersection-complex constituent I H and punctual constituent R Σ H may be described as rigid and flexible in their respective geometric roles; this terminology does not assert a split direct sum or identify either constituent with a KKPY spectral factor. Section 3 also constructs the canonical recollement atom using ϵ can . When the corrected local classes are nonzero, the comparison is ϵ Σ = ϵ can ∘ D λ with fixed quotient markings; the scalar entries of D λ are not fixed to 1. The ( − 2 ) -twisted punctual quotient is different from the ( − 1 ) -twisted term in the normalized nearby/vanishing-cycle interface; Section 2 fixes their functorial and Tate conventions. The latter term also appears with a definite geometric meaning in the Verdier-dual description of the canonical conifold extension in Section 4, rather than being substituted for R Σ H .
At an ordinary double point, H 1 ( i p ! I H ) ≃ Q H ( − 2 ) , while the corresponding degree-one costalk at a smooth point of a threefold vanishes. The resulting local group
Hom D b MHM ( X 0 ) ( i p * Q H ( − 2 ) , I H [ 1 ] )
is one-dimensional. Nevertheless, its dimension does not determine the corrected variation-cone scalar. Recollement constructs the canonical middle term P can H = H 0 ( j * j * I H ) and a particular boundary ϵ can whose costalk coordinate is − 1 under the specified cone rotation. The adjunction-induced comparison P H → P can H has a scalar quotient component λ p at each node; its invertibility follows for a nonzero corrected local class, whereas identity of the two marked extensions requires the additional equality λ p = 1 . That last equality is not claimed here.

1.2. Geometric Input from the Preceding Papers

The relevant antecedents have different bases and coefficients. The projective G L 2 Schubert-convolution calculation of [3] uses pure Hodge modules; its relative affine collision is a separate mixed-coefficient calculation, and its G L 3 incidence tower X 2 → r X 1 → q X 0 supplies actual comparison maps and an octahedral composition defect. The vanishing of a composition defect does not force its constituent coefficient extensions to split. The one-step determinant-conifold construction of [4] uses a framed Higgs evaluation, the invariant-line incidence, an admissible target morphism, and specified star and proper-support operations. Its localized survival results are imported only for their stated charts, coefficients, and opens. The two-fold calculation here retains the intermediate lattice and the first-jet dependence of the transformed Higgs field, rather than replacing the genuine Higgs correspondence by ordinary Schubert convolution.
The determinantal matrix-factorization spectrum, irregular Fourier–Laplace/Stokes data, and stabilized A 2 family of [5] supply independently computed spectral tests, but their underlying coefficient categories are not interchangeable. In particular, the kernel-line small resolution of the conifold is not the full invariant-line incidence, even though both have a projective-line fiber over the zero matrix. In the present work the exceptional curves enter additionally through a proved cohomological Gysin comparison under a projective small-resolution hypothesis; that comparison does not construct a Satake-to-Higgs or Stokes-to-MHM functor.

1.3. Categorical Tools and the Proved Results

Saito’s mixed-Hodge-module formalism supplies the perverse heart, strict filtered differential-module components, nearby and vanishing cycles, and derived operations [6,7,8,9]. Its rational realization is exact and faithful on the hearts but is not generally full. Stable enhancements and coherent six-operation exchange transformations, as used here for finite-type algebraic schemes, are provided by the relevant Saito–Tubach framework [10]. Fedorov’s Hodge–Satake convolution belongs to the equivariant affine-Grassmannian setting, with its specified Tate normalization; the diagrammatic mixed-perverse Hecke categories of Achar–Riche–Vay constitute another coefficient theory [11,12]. None of these results, without an independently constructed kernel comparison, identifies a spherical convolution operator with the actual modified-Higgs operator.
The first principal result concerns the local specialization of the newly defined singular atom. For g = x w − y v in a smooth ambient fourfold, the filtered localization quotient ( O Y [ g − 1 ] / O Y , F ) ( 1 ) realizes the canonical conifold middle module. The Hodge-ideal/minimal-exponent calculation identifies the first relevant filtration levels, and the associated-graded degree-two quotient has no nonzero element annihilated by the maximal ideal. This excludes a graded splitting and proves that the Rees connecting morphism survives z = 0 , both for the canonical extension and, after a nonzero quotient rescaling, for the corrected local extension. The statement is an actual nonvanishing theorem for this filtered class, not a general assertion that Rees specialization is faithful on all derived Hom groups.
The second principal result identifies the global relation morphism of the marked singular atom, under a precise geometric hypothesis. If the nodal projective threefold admits a projective small resolution ρ : X ˜ 0 → X 0 with smooth source and exceptional curves C p , smallness identifies I H 4 ( X 0 , Q ) with H 4 ( X ˜ 0 , Q ) . The adjunction unit for C p , codimension-two purity and Gysin direct image construct a morphism with costalk coordinate + 1 and image PD [ C p ] . The canonical recollement rotation has coordinate − 1 , so for G ρ ( e p ) = PD [ C p ] and the diagonal corrected marking D λ , Section 3 proves
θ ρ ∂ Σ , can H = − G ρ , L Σ , corr H = D λ − 1 ker G ρ .
This identifies the rational relation space, not an integral saturation or torsion lattice, and it does not fix the independent marked scalar λ p . It sharpens the connection with the exceptional-curve relation matrices in the conifold-transition work of Lee–Lin–Wang [13]: a relation matrix has image in ker G ρ , whereas the transpose occurring in their cohomological sequence has a different source and target. Neither map is automatically a quantum F-bundle comparison.
The third geometric result concerns two successive lower minuscule modifications transporting the specified singular-atom arrow. The unrestricted framed lattice-chain surface is F 2 , but Higgs preservation imposes a scheme-theoretic first-jet equation. Section 6 computes the complete punctual incidence as a generically doubled section together with vertical curves over the eigenlines of the first jet, including its scalar exception; the reductions yield rational-curve trees or the full Hirzebruch surface. We compute their rational mixed Hodge cohomology and the terminal-lattice direct images without identifying them with the genuine twice-modified target coefficients. On the latter, the composed Hecke operator preserves the specified connecting arrow under the stated Beck–Chevalley and projection-formula hypotheses. An extraordinary-costalk theorem further proves nonvanishing under explicit smoothness, properness, constant-kernel and node-factorization assumptions. Whether those hypotheses hold on the selected Higgs charts and which numerical perverse boundary maps are nonzero remain separate calculations; the strictly punctual ordinary stalk, by contrast, annihilates the forward arrow.

1.4. The Categorical Atom Comparison

KKPY work with F-bundles over a prescribed non-archimedean analytic base, whose horizontal morphisms and strict exact structure provide categorical representatives for selected spectral factors [1]. Maximality, spectral separation, parity and equivariance requirements must be checked before treating one such representative as an atomic factor. The atom invariant obtained from equivalence classes of factors does not itself carry a canonical Yoneda extension category. A comparison of the conifold arrow therefore has to take place between independently constructed objects and morphisms, not by matching a local spectral polynomial or the number of projective lines. A proposed extension-sensitive realization has the type
R at : C π ⟶ D ex b Fbun k ( B F ) ,
where C π is a geometrically specified source containing the attachment. This display is a required source–target specification, not a functor already constructed.
An actual restricted construction is available on Griffiths-transverse filtered flat bundles ( E , ∇ , F • ) on a smooth base. The exact functor R u sends such an object and its horizontal filtration-preserving morphisms to a formal F-bundle; its special fiber has base-direction operator induced by Griffiths transversality and vanishing leading Euler operator κ = 0 . A single scalar exponential shift changes κ by a scalar, whereas the quantum operator is κ Q = − ( Eu ★ − ) in the specified KKPY convention. For the ordinary non-equivariant graded small-quantum locus of a smooth projective Calabi–Yau variety, the standard Euler vector field vanishes throughout H 2 ( Y ) and hence κ Q = 0 throughout that locus; this particular obstruction disappears, but full connection compatibility and the marked extension square remain unproved. The algebraic z-Rees module of the singular conifold is not an ambient nonsingular flat bundle, although a broader integrable-Rees target permitting singular support is not thereby excluded. Likewise, j * ϵ Σ = 0 for j : X 0 ∖ Σ ↪ X 0 , so no realization factoring solely through this central-fiber smooth-locus restriction can detect the punctual class. Nearby/vanishing-cycle constructions on the total family are different operations and are not ruled out by this vanishing.

1.5. Operations, Hypotheses, and Proof Status

For each finite-type algebraic scheme X, let D H ( X ) denote the stable enhancement of D b MHM ( X ) used in Section 2.2. A marked attachment is an actual arrow α : C → A [ 1 ] ; when convenient the existing arrow category Fun ( Δ 1 , D H ( X ) ) records its homotopy-coherent naturality squares, without creating another realization category. For a framed Higgs correspondence
M − ← p − H μ → q + M + ,
and a specified coefficient kernel including shifts and Tate twists, the relevant star operation is
H μ H ( M ) = q + * K μ ⊗ L p − * M .
It acts on the evaluation pullback of the determinantal conifold attachment, not directly on an arrow living on a different space. The proper-support version with q ! p * is separately normalized; it is not silently replaced by q ! p ! . For two steps, the identified fiber product, the actual kernel, proper base change where required, and the projection formula give a natural equivalence on the selected arrow. Such exactness does not imply nonvanishing or perverse t-exactness.
The filtered operation begins instead with a smooth ambient embedding X 0 ↪ Y and the strictly exact filtered D Y -module image of (1). Its Rees extension and derived specialization use gr D Y ≃ Sym O Y T Y and the algebraic parameter z, not the auxiliary u of the formal F-bundle. In addition to preserving these types, each comparison below specifies the actual map, its categorical domain and codomain, its necessary geometric hypotheses, and the effect of applying it to the fixed connecting arrow. A projective small resolution is invoked only for the exceptional-curve theorem, and the nonvanishing criterion for two-step transport is not applied without its additional target geometry and kernel hypotheses.

1.6. Restricted Mathematical Objectives and Established Outcomes

For subsequent bivariant applications we retain the eight named calculations, now assembled in Theorem 7 under their separate hypotheses. Their established results must remain distinct from outstanding normalization and target-morphism problems:
(R1)
Canonical gluing: open–closed recollement constructs P can H , its canonical attaching morphism, and the map from the corrected middle term; identity of the two marked extensions is reduced to the quotient normalization.
(R2)
Local normalization: the ODP extension group and canonical costalk coordinate − 1 are calculated. The corrected variation-cone comparison scalar λ p is nonzero but remains numerically undetermined.
(R3)
Filtered specialization: the conifold localization and Hodge levels prove nonsplitting of the associated-graded extension at z = 0 ; the Koszul and Bockstein calculations provide independent chain-level descriptions.
(R4)
One-step Hecke transport: the framed evaluation, invariant-line components, genuine target morphism, and normalized pull–tensor–push operations are distinguished; the established localized survival statements are imported within their stated domains.
(R5)
Two-step Higgs geometry: the entire first-jet incidence, doubled section, reduced rational-curve configurations, constant-coefficient images, and bounded properness are computed; identification with the full target requires its actual moduli construction.
(R6)
Two-step attachment: internal MHM composition and the extraordinary costalk formula are established; a smooth-purity theorem supplies sufficient nonvanishing hypotheses, while their verification on the actual Higgs charts and numerical perverse boundaries remain open.
(R7)
Multi-node relations: global connecting maps and rational diagonal rescaling are constructed; when a projective small resolution with smooth source exists, the canonical boundary is the signed exceptional-curve Gysin map and the corrected relation space is the rescaled kernel of the curve-class map.
(R8)
Restricted coherence: the Beck–Chevalley and projection-formula maps, their naturality on the arrow, and explicit octahedral comparison are verified; independently defined cross-category Hecke coherence still requires its own kernel comparison.
The further labels F1–F4 refer to distinct comparison tasks: an independently defined attachment realization, its Rees-specialization image, a comparison with an independently defined Hecke action, and coherence for two compositions. The local specialization question is settled for the nonzero determinantal conifold class, not for arbitrary filtered extensions. The constructed KKPY filtered-flat functor has a different source category and does not settle the proposed singular-to-quantum comparison. This accounting distinguishes proved morphism-level results from restricted geometric applications and from functors that have not been constructed.

1.7. Bivariant Continuation and Organization

Theorem 7 describes the three proved operations on the marked singular conifold Hodge atom. The extension-sensitive data available for a subsequent bivariant theory consist of its fixed connecting morphism, its recollement and small-resolution Gysin comparisons, strict filtered and nonvanishing Rees representatives, the framed one- and two-step Hecke correspondences with their composition maps, and rational global node-relation exact sequences. The two-step costalk theorem supplies a concrete test for a genuine Higgs transport calculation; it does not replace that calculation or provide the numerical perverse boundary maps. Other singular strata, iterated Hecke–Schober gluing, tropical relation transport, and additional coefficient comparisons require independently constructed correspondences and morphisms. In particular, no statement here identifies the type-IIB conifold light sector with a twisted N = 4 magnetic line-operator category. The analytic scattering comparison of [14] and the physical framework of [15] motivate a carefully delimited magnetic interface, but they do not themselves define a functor on the corrected MHM attachment.

1.7.0.1. Organization.

Section 2 fixes the geometric setting, coefficients, t-structures, and functorial conventions. Section 3 defines the marked singular conifold Hodge atom and constructs canonical gluing, the marked local comparison, and the exceptional-curve relation theorem. Section 4 proves conifold Rees non-splitting and records the filtered realizations. Section 5 and Section 6 treat framed one-step transport, the two-step Higgs first-jet incidence, coherent convolution, and the extraordinary-costalk nonvanishing criterion. Section 7 develops categorical F-bundle representatives and the exact filtered-flat Rees functor, explicitly separating this construction from a singular-to-quantum comparison. Section 8 treats the restricted magnetic Hecke interface; Section 9 proves the synthesis theorem for the marked singular atom and records the remaining hypotheses and bivariant continuation. Appendices Appendix A–Appendix F give the costalk and Gysin calculations, Rees complexes, complete Hecke-lattice and costalk models, Stokes tests, magnetic admissibility conditions, and formal F-bundle computations.

2. Categories, Coefficients, and the Attachment-Bearing Test Category

We fix the geometric bases, exact and derived coefficient categories, and typed morphisms used in both mathematical tasks. The conifold attachment belongs to D b MHM ( X 0 ) ; the filtered component is a functor into an exact category on a chosen smooth ambient variety; and KKPY atomic representatives belong to a separate, non-archimedean category of F-bundles. These data are not related by an unnamed realization operation. We use the notation and conventions of this section in subsequent proofs, where each comparison must identify both its categorical source and its geometric base.

2.1. Geometric Base and Cohomological Conventions

Definition 1
(Geometric setting). Let B be a smooth connected complex algebraic curve, let 0 ∈ B ( C ) , and let
π alg : X alg ⟶ B
be a proper algebraic morphism. Define thealgebraiccentral fiber X 0 : = π alg − 1 ( 0 ) . Let Δ ⊂ B an be a sufficiently small analytic disk about 0 and set
X = X alg an × B an Δ , π : X ⟶ Δ .
For t ∈ Δ denote the analytic fiber by X t an = π − 1 ( t ) ; in particular X 0 an is the analytification of the algebraic scheme X 0 . Assume that:
  • X t an is smooth of complex dimension three for t ≠ 0 ;
  • X 0 is reduced and equidimensional of complex dimension three;
  • Sing ( X 0 ) = Σ = { p 1 , … , p N } is finite;
  • every p ∈ Σ is an ordinary double point.
No global small resolution of X 0 is assumed. Unless an analytification superscript is displayed, MHM ( X 0 ) and its derived category refer to thealgebraiccentral fiber.
The algebraic morphism π alg supplies the base for Saito’s algebraic mixed-Hodge-module operations. The analytic disk is used for local monodromy, Milnor fibers and analytic normal forms. When a local analytic calculation is invoked in an algebraic MHM statement, we specify the analytification/comparison step or the algebraic chart to which it applies. Local analytic existence does not itself produce an algebraic global coefficient object.
For an ordinary double point p, there are analytic coordinates ( x , y , z , w ) in a neighborhood of p in a smooth fourfold such that the germ of the central fiber is
( X 0 , p ) ≃ { x w − y z = 0 } , 0 .
This equation describes the central-fiber singularity, not the total space of the degeneration. A local smoothing may be given by x w − y z = t only under the corresponding local-equivalence hypothesis.
Set U = X 0 ∖ Σ and let j : U ↪ X 0 and i p : { p } ↪ X 0 be the inclusions. These strata, or a Whitney refinement compatible with the maps, are used for constructibility arguments.
Definition 2
(Coefficient categories). For a complex algebraic variety Z, let D c b ( Z , Q ) be the bounded derived category of algebraically constructible rational sheaf complexes, and let Perv ( Z , Q ) be the heart of its middle-perversity t-structure. Let MHM ( Z ) denote Saito’s abelian category of graded-polarizable rational mixed Hodge modules, and put D b MHM ( Z ) = D b ( MHM ( Z ) ) . We write
rat Z : MHM ( Z ) ⟶ Perv ( Z , Q )
for the exact faithful rational functor and use the same symbol for its derived extension to D c b ( Z , Q ) .
These categories are not interchangeable. An object of MHM ( Z ) includes a rational perverse sheaf and compatible filtered regular holonomic differential-module and weight-theoretic structures. Locally on a smooth ambient variety its data may be represented schematically by
( M , F • M ) , K Q , α , W • ,
where M is the differential module, K Q the rational perverse sheaf, α the de Rham comparison, and W • the weight filtration. Compatibility, specializability, admissibility, and relative-monodromy conditions restrict these data. An arbitrary filtered regular holonomic differential module with a rational perverse-sheaf structure need not define a mixed Hodge module [6,8].
An object of D b MHM ( Z ) is represented by a bounded complex
… ⟶ M k − 1 → d k − 1 M k → d k M k + 1 ⟶ …
whose terms and differentials belong to MHM ( Z ) , up to quasi-isomorphism. Its cohomology objects are themselves mixed Hodge modules. The derived rational functor satisfies
rat Z H MHM k ( M • ) ≃ H k p rat Z ( M • ) .
Ordinary cohomology sheaves need not coincide with perverse cohomology.
Definition 3
(Shift and truncation conventions). For a cochain complex K • , the shift is defined by K [ 1 ] n = K n + 1 and d K [ 1 ] n = − d K n + 1 . For a : A → B , the cone has terms
Cone ( a ) n = B n ⊕ A n + 1 , d Cone ( a ) = d B a 0 − d A .
Its distinguished triangle is A → a B → Cone ( a ) → A [ 1 ] . We denote standard truncations by τ ≤ k , τ ≥ k , and middle-perversity truncations by τ ≤ k p , τ ≥ k p . The standard heart of D b MHM ( Z ) is MHM ( Z ) , whose rational images are perverse sheaves.
Definition 4
(Intersection-complex normalization). Let I H = IC X 0 H . We normalize the underlying rational perverse sheaf by
rat X 0 ( I H ) = j ! * Q U [ 3 ] .
The intermediate extension is taken in the abelian perverse category.
The shift [ 3 ] is fixed by the complex dimension of U, not the ambient dimension four.
Definition 5
(Tate-twist conventions). For m ∈ Z , let Q H ( m ) denote the Tate Hodge structure of weight − 2 m , regarded as a mixed Hodge module on a point. In particular,
wt Q H ( − 1 ) = 2 , wt Q H ( − 2 ) = 4 .
For M ∈ MHM ( Z ) , let M ( m ) denote its Tate twist. Tate twists are not cohomological shifts.
The exact functor i p * : MHM ( { p } ) → MHM ( X 0 ) defines the punctual forward quotient
R Σ H = ⨁ p ∈ Σ i p * Q H ( − 2 ) .
It lies in perverse degree zero. We write f * , f ! , f * , f ! for the derived mixed-Hodge-module operations, suppressing derived symbols when unambiguous. The stalk and costalk i p * M , i p ! M are complexes of mixed Hodge structures. For an open immersion, j * ≃ j ! , while j ! and j * generally differ. For a proper morphism f, f ! ≃ f * , but this does not permit replacement of nonproper direct image by proper-support image.
Proposition 1
(Smooth-point contrast for the punctual coefficient). Let T be a smooth complex algebraic threefold, x ∈ T a closed point, and k : { x } ↪ T . Put I T H = Q T H [ 3 ] , the perverse-normalized constant Hodge module. Then
Hom D b MHM ( T ) k * Q H ( − 2 ) , I T H [ 1 ] = 0 .
Proof. 
Absolute purity and the chosen left/perverse normalization give
k ! I T H ≃ Q H ( − 3 ) [ − 3 ] in D b MHM ( { x } ) = D b MHS Q .
By the adjunction k * ⊣ k ! , the displayed Hom group equals
Hom D b MHS Q Q H ( − 2 ) , Q H ( − 3 ) [ − 2 ] .
The source belongs to the standard heart and the target to D ≥ 2 , so this group vanishes by the standard t-structure orthogonality. For an ordinary double point, by contrast, the degree-one costalk term computed in Appendix A permits the nonzero local extension line; that calculation is not inferred from purity on a smooth space. □
Remark 1
(Small resolutions). A projective small resolution ρ : X ˜ 0 → X 0 will be assumed only when explicitly required. Its existence is not implied by the isolated ordinary-double-point hypothesis. Exceptional curves alone do not identify an extension class: a boundary or Gysin morphism and its normalization must be specified.

2.2. Enhanced Operations and Hecke Coefficient Categories

We distinguish an established enhanced category from a proposed attachment-preserving action. Let X be a separated finite-type scheme over C . We write
D H ( X ) = D ∞ b MHM ( X )
for the bounded stable ∞-categorical enhancement of Saito’s derived category furnished by the compatible six-functor framework of [10]. The symbol D ∞ b records the enhancement, while its homotopy category is D b MHM ( X ) . For a morphism f : X → Y of the type allowed in this formalism, the operations f * , f ! , f * , f ! act as exact functors of stable categories with the established exchange morphisms. They take any arrow a : C → A [ 1 ] to the corresponding derived arrow, but exactness does not assert that its image is nonzero.
For a connected reductive group G over C , set O = C [ [ t ] ] , K = C ( ( t ) ) , and Gr G = G ( K ) / G ( O ) . Fedorov’s equivariant Hodge–Satake category is built from G ( O ) -equivariant Hodge modules on Gr G . For suitable equivariant objects K , L the twisted exterior product K ⊠ ˜ L is defined on G ( K ) × G ( O ) Gr G by finite-stage descent. With m denoting the ind-proper convolution map, the pure heart carries the product
K ★ L = H 0 m * ( K ⊠ ˜ L ) .
Its perverse concentration, associativity, and compatibility with the rational perverse-sheaf convolution are established in [11]. The mixed version is stated in [11]; we do not infer additional derived or mixed convolution results beyond those established there.
For a Coxeter system ( W , S ) with an admissible realization, let H diag ( W ) denote the monoidal diagrammatic category of Elias–Williamson type used by [12]. Its biequivariant bounded homotopy category carries a convolution ★ and recollement along Coxeter strata. The standard and costandard objects Δ w , ∇ w belong to its mixed-perverse framework, and the connecting morphisms in the convolution triangles of [12] are actual derived morphisms. In this setting [ 1 ] is cohomological translation and 〈 1 〉 is the diagrammatic grading operation; neither is identified with the Tate twist Q H ( 1 ) without a separate comparison.
These three coefficient systems are not identified here. The enhanced MHM category provides the ambient target for the conifold extension; the spherical category provides a candidate source of Hecke kernels when the relevant geometric descent has been constructed; and the diagrammatic category supplies a distinct model for stratified gluing and convolution. The existence of a monoidal action, or of a functor relating these categories on a chosen conifold chart, is an additional assertion requiring an explicitly defined kernel and exchange transformations.

2.3. The Filtered-Component Functor and Its Exact Target

Let ι : X 0 ↪ Y be a fixed closed embedding into a smooth complex algebraic variety on the chart under consideration, and let D Y denote the sheaf of left algebraic differential operators with its increasing order filtration F ord . For a general singular variety admitting only local embeddings, the construction is applied on compatible smooth ambient charts; a global D X 0 in the smooth-variety sense is not assumed.
Definition 6
(Good filtered regular-holonomic category). Let FilRH ( Y ) be the category whose objects are pairs ( M , F • M ) such that M is a regular-holonomic left D Y -module and F • M is a good exhaustive increasing filtration with
F a ord D Y · F b M ⊆ F a + b M .
A morphism f : ( M , F ) → ( N , G ) is a D Y -linear map satisfying f ( F p M ) ⊆ G p N for every p ∈ Z . A composable pair of such morphisms has filtration-preserving composite, and identities preserve all filtrations. We equip this category with its strict exact structure: a short sequence is a conflation when its underlying module sequence and the sequence at every filtration index p are exact. Write FilRH str ( Y ) when emphasizing this exact structure. The notation does not assert that every morphism is strict.
Proposition 2
(Filtered component as an exact functor). For the fixed embedding ι there is an exact functor
For F , ι : MHM ( X 0 ) ⟶ FilRH str ( Y ) .
It sends M to the good filtered left D Y -module component of ι * M and a morphism to its filtered D Y -linear component. Its target exactness is strict exactness, not an assertion that FilRH ( Y ) is abelian.
Proof. 
For a closed embedding, the exact fully faithful direct image identifies MHM ( X 0 ) with the Hodge modules on Y supported on X 0 . By Saito’s definition of a morphism of mixed Hodge modules, the filtered differential-module component of a morphism preserves the filtration; its identity and composite are the corresponding identity and composite of the filtered components. This proves the functor axioms [6,8].
Let 0 → A → u P → v R → 0 be short exact in MHM ( X 0 ) . Write ( A , F ) , ( P , F ) , ( R , F ) for its filtered components after ι * . Exactness of the regular-holonomic component gives 0 → A → u F P → v F R → 0 . The strictness theorem for MHM morphisms gives, for every p,
u F ( A ) ∩ F p P = u F ( F p A ) , v F ( F p P ) = v F ( P ) ∩ F p R = F p R .
If m ∈ F p P and v F ( m ) = 0 , underlying exactness gives m = u F ( a ) for some a ∈ A . The first strictness identity then gives a ∈ F p A , because u F is injective. Thus
0 ⟶ F p A ⟶ F p P ⟶ F p R ⟶ 0
is exact for every p. Hence the filtered-component sequence is a conflation and the functor is exact in the stated sense [9]. □
Remark 2
(No comparison to filtered flat bundles yet). The object For F , ι ( M ) is generally a filtered regular-holonomic module supported on a singular locus in Y; it need not be a vector bundle with a flat connection on a smooth quantum-parameter base. Passing to its Rees module is a well-defined filtered-module construction, but does not provide the input type for a filtered-flat-to-TE functor without an additional geometric operation.

2.4. KKPY F-bundles and Spectral Categorical Carriers

The KKPY definitions are stated in a non-archimedean analytic category distinct from MHM ( X 0 ) . We fix this target before formulating any map to it. In particular, no scalar extension from Q or C to the chosen non-archimedean field, or identification of the two parameter bases, is made implicitly.
Definition 7
(KKPY F-bundle category). Fix a complete non-archimedean normed field k of characteristic zero, a smooth k -analytic super-base B F (or an admissible germ), and a disk germ D u with coordinate u. Following [1], the category
Fbun k ( B F )
has objects ( H , ∇ ) / B F , where H is an analytic super vector bundle on B F × D u and ∇ is flat and meromorphic with poles at most along u = 0 , such that ∇ u ξ and ∇ u 2 ∂ u preserve the analytic lattice for every local vector field ξ on B F . A morphism is an even bundle morphism on this fixed base that intertwines the two meromorphic connections. Identities and composition are those of horizontal bundle morphisms. KKPY show that this is a k -linear additive, Karoubi-closed, quasi-abelian category; its exact structure is the canonical strict exact structure.
The atom invariant is not this category. In KKPY an atomic F-bundle is a G-equivariant germ satisfying maximality, compatibility of the designated parity element with the super grading, and a single eigenvalue for the specified Euler operator on the invariant locus [1]. A global atom is formed by further equivalences among geometric atomic representatives [1]; it does not acquire Hom sets by being an equivalence class.
Definition 8
(Spectrally marked carrier). Fix a common analytic base B F to which all relevant spectral factors have been pulled back by specified maps. Define the category
SpecMarkFbun k ( B F ) .
Its objects are pairs ( F , e ) , where F is an object of Fbun k ( B F ) and e ∈ End ( F ) is a horizontal idempotent with e 2 = e . The idempotent must come from aspecified decomposition of F-bundlesin the regime of the spectral theorem [1]. Morphisms h : ( F , e ) → ( F ′ , e ′ ) are F-bundle morphisms satisfying
h ∘ e = e ′ ∘ h .
A spectral marking at u = 0 alone is not taken to be a horizontal projector unless it extends to such a decomposition.
Lemma 1
(The marked carrier is a category). The objects and morphisms of Definition 8 form a category, with the forgetful functor to Fbun k ( B F ) .
Proof. 
The identity of F satisfies id F e = e id F . If h e = e ′ h and k e ′ = e ′ ′ k , then
( k ∘ h ) e = k ( e ′ h ) = ( k e ′ ) h = e ′ ′ ( k ∘ h ) .
Thus composition is closed. Associativity follows from that in Fbun k ( B F ) , and forgetting the idempotent preserves identities and composition. □
Remark 3
(Exactness is not inherited automatically). The spectrally marked collection is not asserted to be extension-closed or quasi-abelian. Yoneda extensions, when required, are taken in the ambient quasi-abelian F-bundle category, or in an explicitly constructed derived category of its exact structure. A chosen idempotent does not by itself satisfy the G-atomic conditions, and an isomorphism of selected factors is not a global equivalence of atom invariants.

2.5. The Corrected Variation-Cone Input

Let f : Y → B be a regular function or morphism on a smooth algebraic ambient space, with central fiber f − 1 ( 0 ) , and let M ∈ MHM ( Y ) satisfy the required specializability hypotheses. Let Ψ f H ( M ) and Φ f H ( M ) denote the perverse-normalized unipotent nearby- and vanishing-cycle objects. If the symbols ψ f , 1 H and ϕ f , 1 H denote the unshifted sheaf-theoretic conventions, our normalization is
Ψ f H ( M ) = ψ f , 1 H ( M ) [ − 1 ] , Φ f H ( M ) = ϕ f , 1 H ( M ) [ − 1 ] .
No additional [ − 1 ] is applied when the source already uses perverse-normalized functors. The canonical and variation maps below always refer to the resulting objects in the selected convention; their rational images are the corresponding perverse-normalized cycle morphisms.
Fix the canonical map and Tate-twisted logarithmic variation map
can H : Ψ f H ( M ) → Φ f H ( M ) , var H : Φ f H ( M ) → Ψ f H ( M ) ( − 1 ) .
Write T u for unipotent monodromy and N for its nilpotent logarithm with Tate-twisted target. With the logarithmic normalization of variation, the relations are
var H ∘ can H = N Ψ , can H ( − 1 ) ∘ var H = N Φ .
With the unmodified topological variation, the corresponding compositions use T u − 1 . The conventions must not be interchanged without tracking the power-series normalization relating them.
The corrected construction in [2] uses a variation cone with its specified shifts and Tate normalization:
V f ( M ) = Cone ( var H ) [ − 1 ] .
This is a derived complex, not an assertion that an arbitrary variation cone lies in the perverse heart. Its identification with the corrected middle term requires the specified gluing, truncation, and comparison maps of the source construction.
Definition 9
(Corrected forward-extension datum). Under the hypotheses of [2], fix ( I H , P H , R Σ H ; u , v ) in MHM ( X 0 ) with the exact sequence
0 ⟶ I H → u P H → v R Σ H ⟶ 0 .
Here I H and R Σ H have the normalizations fixed above.
This extension determines
ϵ Σ : R Σ H ⟶ I H [ 1 ]
in D b MHM ( X 0 ) , and the distinguished triangle
I H → u P H → v R Σ H → ϵ Σ I H [ 1 ] .
Proposition 3
(Derived representation of a forward extension). Let 0 → A → a P → b C → 0 be a short exact sequence in MHM ( X 0 ) with Yoneda connecting morphism ϵ : C → A [ 1 ] . Then A → P → C → ϵ A [ 1 ] is distinguished and
P ≃ Cone ( ϵ ) [ − 1 ]
compatibly with its structure maps.
Proof. 
Regard A , P , C as complexes in degree zero. The exact sequence defines an element of Ext MHM ( X 0 ) 1 ( C , A ) , identified with Hom D b MHM ( X 0 ) ( C , A [ 1 ] ) . The associated distinguished triangle is A → P → C → ϵ A [ 1 ] . Rotating gives C [ − 1 ] → ϵ [ − 1 ] A → P → C . By the cone convention, P ≃ Cone ( ϵ [ − 1 ] ) ≃ Cone ( ϵ ) [ − 1 ] . The triangle identification respects the original extension maps. □
Remark 4
(Two distinct punctual contributions). For clarity, write the two punctual Tate models separately:
W Σ H : = ⨁ p ∈ Σ i p * Q H ( − 1 ) , R Σ H : = ⨁ p ∈ Σ i p * Q H ( − 2 ) .
The first symbol records the ( − 1 ) normalization used for the punctual model in the local nearby/vanishing-cycle interface; it is not the entire nearby- or vanishing-cycle complex. The second is the forward quotient of (14). They have different weights. Any comparison requires a constructed boundary, costalk, or Gysin map with its specified cohomological shift and Tate normalization.
Definition 10
(Local calibration data). For p ∈ Σ , the local calibration is the collection
i p * I H , i p ! I H , i p * P H , i p ! P H , i p * R Σ H , i p ! R Σ H ,
and all maps induced by u , v , ϵ Σ , recording each nonzero cohomological degree, Hodge weight, and Tate twist.
Since the other summands are supported away from p,
i p * R Σ H ≃ Q H ( − 2 ) , i p ! R Σ H ≃ Q H ( − 2 ) .
Applying i p * and i p ! to the attachment triangle gives
i p * I H → i p * P H → Q H ( − 2 ) → i p * ϵ Σ i p * I H [ 1 ] ,
and
i p ! I H → i p ! P H → Q H ( − 2 ) → i p ! ϵ Σ i p ! I H [ 1 ] .
Their long exact sequences are taken in mixed Hodge structures. These formal triangles do not supply the numerical stalk and costalk calculation. Appendix A must determine the relevant groups, filtrations, maps, and signs, and calibrate the variation cone against both triangles.

2.6. Marked Arrows and the Scope of the Test Objects

For the attachment calculations in this paper, the ambient category is already D H ( X 0 ) ; no additional category needs to be constructed in order to evaluate the established exact functors on the specified arrow. A marked-arrow category is useful only when several independently constructed attachment diagrams must be compared functorially.
Definition 11
(Enhanced derived category). Let D H ( X ) denote the stable enhancement of D b MHM ( X ) reviewed in Section 2.2. Its mapping spaces satisfy
π 0 Map D H ( X ) ( M , N ) ≃ Hom D b MHM ( X ) ( M , N ) .
Definition 12
(Marked attachment). An attachment on X is a specified morphism
α : C ⟶ A [ 1 ]
in D H ( X ) , together with its geometrically defined endpoint markings and the corresponding fiber sequence
A ⟶ P α ⟶ C → α A [ 1 ] , P α ≃ Cone ( α ) [ − 1 ] .
A morphism to another marked attachment β : C ′ → A ′ [ 1 ] is a homotopy-coherent square with endpoint maps f C : C → C ′ and f A : A → A ′ , respecting the markings. In the homotopy category it satisfies
f A [ 1 ] ∘ α = β ∘ f C .
The associated arrow category is the existing stable category
Arr ( D H ( X ) ) = Fun ( Δ 1 , D H ( X ) ) .
Our designated object is
e Σ = ( R Σ H → ϵ Σ I H [ 1 ] ) .
Its geometric markings are those supplied by the corrected extension, not chosen after applying a proposed comparison functor.
Proposition 4
(Exact functors preserve attachment triangles). Let F : D → E be an exact functor of stable categories and let α : C → A [ 1 ] be an arrow. Then F induces an exact functor on arrow categories and a natural equivalence
F ( P α ) ≃ P F ( α ) .
Proof. 
Applying F to the fiber sequence P α → C → α A [ 1 ] gives a fiber sequence F ( P α ) → F ( C ) → F ( α ) F ( A ) [ 1 ] , because an exact functor preserves finite limits and suspension. The universal property of the fiber identifies its first term with P F ( α ) . The construction is natural on homotopy-coherent squares, hence gives the asserted arrow functor. □
Definition 13
(Detection and identification of an attachment). An exact functor Fdetectsα when F ( α ) ≠ 0 in the target homotopy category. For a separately specified target arrow ϵ : R → I [ 1 ] , anidentificationconsists of endpoint isomorphisms η C : F ( C ) → ∼ R and η A : F ( A ) → ∼ I with
η A [ 1 ] ∘ F ( α ) = ϵ ∘ η C ,
and a coherent commutativity datum when an enhanced statement is required.
Remark 5
(Nonvanishing is not formal). Proposition 4 does not show F ( α ) ≠ 0 . In particular, the functor forgetting a mixed Hodge structure can carry a nonsplit mixed-Hodge extension to a split extension of rational vector spaces. Detection and identification must therefore be proved for the chosen morphism.
Remark 6
(Internal closure and the geometric domain). If later composition arguments require a category of several marked attachments, one may use the full subcategory of the existing arrow category on the geometric objects under consideration, or its stable idempotent-complete closure where appropriate. Such closure does not establish that a Hecke functor preserves the geometric class of objects. The latter requires a separate check of the correspondence and coefficients. We do not postulate a universal comparison from matrix factorizations, irregular Stokes categories, or diagrammatic Hecke categories to D H ( X 0 ) .
Remark 7
(Soundness and scope). For an independently constructed comparison functor,soundnessmeans that the prescribed arrow square (21) commutes. Thescopeis the explicitly verified class of geometric spaces, coefficients, and correspondences on which the functor is defined. Neither follows from the other. An admissible source category will be introduced only when a later theorem requires and constructs one independently of the desired target class.

2.7. Morphisms, Realizations, and the Smooth-Locus Obstruction

A morphism f : X → Y of algebraic schemes, its derived pullback f * , a morphism of Hodge modules α : M → N [ 1 ] , and a natural transformation between functors are four different kinds of data. We use the word realization only after specifying a genuine categorical operation on both objects and morphisms. A comparison of independently defined objects in different categories requires functors to a fixed common target and a morphism there; the existence of a common numerical invariant does not construct that morphism.
The projection For F , ι in Proposition 2 has target FilRH str ( Y ) . A second, independent starting category for a formal F-bundle construction is the category of finite-rank filtered flat bundles ( E , ∇ B , F • E ) on a smooth base B ♭ with filtration by subbundles, locally free successive quotients, and Griffiths transversality ∇ B ( F p E ) ⊆ F p − 1 E ⊗ Ω B ♭ 1 . Morphisms are horizontal filtration-preserving bundle maps. Its formal Rees-to-TE construction is developed separately; it is not a functor with source FilRH str ( Y ) as presently defined. Neither the equality of the two bases nor extension of scalars to the chosen non-archimedean coefficient field is implicit.
Proposition 5
(Smooth-locus factorization annihilates the punctual attachment). For the open immersion j : U = X 0 ∖ Σ ↪ X 0 and the specified attaching morphism (15), one has
j * ϵ Σ = 0 in Hom D b MHM ( U ) ( j * R Σ H , j * I H [ 1 ] ) .
Consequently, if T : D b MHM ( U ) → E is any exact functor to a stable category, then ( T ∘ j * ) ( ϵ Σ ) = 0 .
Proof. 
For each p ∈ Σ , the Cartesian intersection U × X 0 { p } is empty. The open–closed support relation gives j * i p * Q H ( − 2 ) = 0 . Since Σ is finite,
j * R Σ H = ⨁ p ∈ Σ j * i p * Q H ( − 2 ) = 0 .
Applying the exact functor j * to ϵ Σ therefore yields a morphism from the zero object to j * I H [ 1 ] , hence the zero morphism. Applying T preserves zero objects and zero morphisms and proves the second assertion. □
Remark 8
(What the obstruction does not say). The preceding proposition concerns restriction to thesmooth locus of the singular central fiber. It is not a statement about the nearby smooth fibers X t , their Gauss–Manin variation, the nearby-cycle functor, or a degeneration comparison involving the total family X . Those constructions have different sources and can retain information not present in j * ϵ Σ . No functor from the marked conifold attachment to a KKPY atomic F-bundle has been constructed in this section.

2.8. The Separate Stabilized A 2 Collision Test

Let S be a small analytic disk with coordinate s and let S × = S ∖ { 0 } . Consider the family
F s ( x , y , z , w ) = x 3 3 − s x + y 2 + z 2 + w 2
on C 4 . Its relative critical locus is defined by d x , y , z , w F s = 0 , equivalently
Crit ( F / S ) = { x 2 = s , y = z = w = 0 } .
Indeed,
∂ F s ∂ x = x 2 − s , ∂ F s ∂ y = 2 y , ∂ F s ∂ z = 2 z , ∂ F s ∂ w = 2 w .
Over S × a chosen local branch of s gives two critical points q + ( s ) = ( s , 0 , 0 , 0 ) and q − ( s ) = ( − s , 0 , 0 , 0 ) . The Hessian is
Hess ( F s ) = 2 x 0 0 0 0 2 0 0 0 0 2 0 0 0 0 2 , det Hess ( F s ) = 16 x .
Thus both points are Morse for s ≠ 0 , while the central critical point has Hessian rank three.
Proposition 6
(Critical-value cover). The critical values of (22) are T ± ( s ) = ∓ 2 3 s 3 / 2 . Their unordered pair defines
T 2 = 4 9 s 3 .
The projection to the s-disk is generically two-to-one, ramified at the origin. Analytic continuation once around the origin interchanges the critical points and critical values.
Proof. 
At a critical point, s = x 2 and y = z = w = 0 . Hence
F s ( x , 0 , 0 , 0 ) = x 3 3 − s x = − 2 3 x 3 .
Substituting x = ± s gives the asserted values; squaring yields T 2 = 4 9 s 3 . The normalization is parametrized by x with s = x 2 and T = − 2 3 x 3 . The map x ↦ x 2 is ramified at zero and two-to-one elsewhere. A loop about s = 0 takes s to − s , interchanging the sheets and their values. □
The critical-value curve is cuspidal. The normalized generic critical-point cover has two sheets, but the central fiber contains a single non-Morse point. Hence the generic spectral decomposition cannot be transported across the collision without specifying its extension and descent data.
Proposition 7
(Generic projector interpolation). Let K = C ( ( x ) ) and R = C [ [ x ] ] , with s = x 2 and T = − 2 3 x 3 . On V = K 2 set
D = − 2 3 x 3 0 0 2 3 x 3 .
The spectral projectors for the two distinct generic eigenvalues are given by
e + = D − 2 3 x 3 I − 4 3 x 3 , e − = D + 2 3 x 3 I 4 3 x 3 .
These generic interpolation expressions have denominators vanishing at x = 0 . This does not, by itself, preclude extension of the projectors on a specially chosen lattice.
Proof. 
The eigenvalues are λ + = − 2 3 x 3 and λ − = 2 3 x 3 . Lagrange interpolation yields e + = ( D − λ − I ) / ( λ + − λ − ) and e − = ( D − λ + I ) / ( λ − − λ + ) . The denominators are ∓ 4 3 x 3 , invertible in K but not R. For the displayed diagonal lattice R 2 , these expressions simplify to constant diagonal idempotents and do extend. Thus an intrinsic nonextension claim requires a fixed lattice, descent condition, or endomorphism algebra. □
Remark 9
(Monodromy and descent of projectors). Monodromy on S × exchanges the labeled critical points. Accordingly, individually labeled projectors do not descend as single-valued labeled objects without accounting for this action. This descent issue differs from extension across s = 0 after normalization, which depends on the chosen lattice.
The local model is related to the irregular Fourier–Laplace and Stokes constructions of [5,16,17]. No morphism between those categories and the corrected MHM attachment category follows merely from critical-value collision. In particular, spectral collision does not by itself imply a nonzero Yoneda extension; projector degeneration does not imply a nonzero Hecke composition defect; and critical-value monodromy does not imply a Langlands obstruction. Each comparison needs a specified functor or natural transformation on objects and maps.
Remark 10
(The two local tests). The ordinary double point supplies the designated derived morphism ϵ Σ : R Σ H → I H [ 1 ] . The stabilized A 2 collision supplies the critical-value family T 2 = 4 9 s 3 and its monodromy. The former tests images of a specified attachment; the latter tests degeneration and descent of spectral data. They remain separate until a comparison functor with specified source, target, objects, morphisms, and coherences is built.

3. Canonical Gluing and Extension-Sensitive mixed-Hodge-module Transport

We use the geometric base, coefficient categories, t-structures, and cone convention of Section 2. This section treats the fixed corrected triangle
I H → u P H → v R Σ H → ϵ Σ I H [ 1 ]
inside D b MHM ( X 0 ) , rather than assuming a new realization category. Definition 15 designates this marked extension-bearing datum as a singular conifold Hodge atom; the designation is internal to mixed Hodge modules and does not identify it with a KKPY spectral factor. There are two distinct tasks. First, the open–closed recollement constructs a canonical middle term and a particular connecting morphism; the subsequent MacPherson–Vilonen comparison tests whether the corrected variation-cone arrow agrees with that morphism with its fixed endpoint markings. Second, an actual evaluation map and Hecke correspondence transport the designated arrow through Saito’s six operations. Neither task requires an unconstructed KKPY F-bundle functor.
The localization identification and the conifold non-splitting theorem of Section 4 provide an explicit positive specialization result; the exact corrected marking remains a separate local problem. For a projective small resolution with smooth source, the boundary–Gysin comparison is constructed in Appendix A: the canonical relation map is the negative exceptional-curve class map under the complex-orientation convention. This geometric identification is distinct from determining the marked variation-cone scalar.
The local ordinary double point is essential: its degree-one intersection-complex costalk produces the punctual Q H ( − 2 ) quotient, whereas the analogous costalk group at a smooth threefold point vanishes. This difference permits a canonical extension that is absent in the smooth local model. We calculate the adjunction boundary, its cohomological sign, and the map comparing it with the corrected middle term. We then distinguish actual derived transport from nonvanishing and from perverse placement on the target. The Schubert, affine-collision, and framed-Higgs inputs of [3,4,5] retain their respective spaces and coefficient categories.

3.1. Provenance of the Three Inputs and Their Coefficient Categories

Definition 14
(Corrected conifold datum). Let X 0 be the algebraic central fiber fixed in Section 2, let Σ ⊂ X 0 be its finite ordinary-double-point locus, and let i p : { p } ↪ X 0 be the closed immersion. Thecorrected conifold datum is the exact sequence in MHM ( X 0 )
0 ⟶ I H → u P H → v R Σ H ⟶ 0 , R Σ H = ⨁ p ∈ Σ i p * Q H ( − 2 ) ,
with its specified connecting morphism ϵ Σ in (25). The maps u , v and the normalization of ϵ Σ form part of the datum.
Definition 15
(Marked singular conifold Hodge atom). For the nodal threefold X 0 , the fixed node set Σ, and the coefficient category MHM ( X 0 ) of Definition 14, amarked singular conifold Hodge atomis the extension-bearing datum
A Σ H : = ( I H , R Σ H , ϵ Σ ; u , v ) , I H = IC X 0 H , R Σ H = ⨁ p ∈ Σ i p * Q H ( − 2 ) ,
where u : I H → P H , v : P H → R Σ H and ϵ Σ : R Σ H → I H [ 1 ] are thespecifiedmaps of the short exact sequence (26) and its derived triangle (25). We call I H thepersistent intersection-complex constituentand R Σ H thepunctual constituent. The termsrigidandflexible, when used for these two constituents, designate these respective geometric roles; they do not assert a direct-sum splitting, a decomposition into independently defined atomic objects, or an identification with a spectral atom in the sense of [1].
Thecanonical recollement atom A Σ , can H is the same type of marked datum obtained from the canonical exact sequence (38) and its connecting morphism ϵ can constructed in Proposition 9. Thecorrected marked atomis A Σ H as defined above, with its original variation-cone normalization. These objects are defined in the existing mixed-Hodge-module and derived categories; the terminology introduces no additional realization category or functor.
Remark 11
(What is intrinsic to the marked atom). The middle term can be recovered, compatibly with the designated triangle, as P H ≃ Cone ( ϵ Σ ) [ − 1 ] . Thus the atom retains the extension morphism rather than merely the two endpoint objects or their classes in a Grothendieck group. When all local corrected classes are nonzero, Proposition 15 identifies ϵ Σ = ϵ can ∘ D λ for the fixed punctual identifications. The associated middle Hodge modules are then isomorphic, but the two marked atoms need not be identity-markedly isomorphic: that requires D λ = id . This comparison does not determine the individual scalars λ p . The termsingular atomrefers here to the specified conifold extension and not to the independently defined KKPY spectral decomposition.
Proposition 8
(Brick criterion and indecomposability of a singular conifold Hodge atom). Assume that X 0 is irreducible and that the corrected conifold datum of Definition 14 has the fixed punctual quotient R Σ H = ⨁ p ∈ Σ i p * Q H ( − 2 ) . Then its middle Hodge module P H is indecomposable in MHM ( X 0 ) if and only if every local attaching component ϵ p : i p * Q H ( − 2 ) → I H [ 1 ] is nonzero. If these components are all nonzero, the stronger conclusion holds:
End MHM ( X 0 ) ( P H ) ≃ Q .
In particular, when every component is nonzero, P H is abrick: its endomorphism ring is the field Q . This criterion concerns the middle object, not equality of the canonical and corrected endpoint markings; it does not determine any of the scalars λ p .
Proof. 
Since X 0 is irreducible, its smooth locus U is connected. The intersection Hodge module I H , obtained by intermediate extension of the rank-one constant variation on U, is simple. Its endomorphisms restrict injectively to endomorphisms of the constant rank-one variation on U, and scalar endomorphisms extend; thus End ( I H ) = Q . It is not isomorphic to any punctual simple object, so Hom MHM ( X 0 ) ( I H , R Σ H ) = 0 . Applying Hom ( I H , − ) to the given short exact sequence therefore identifies
Hom MHM ( X 0 ) ( I H , P H ) ≃ End MHM ( X 0 ) ( I H ) = Q ,
with generator the specified inclusion u : I H ↪ P H .
Apply Hom MHM ( X 0 ) ( R Σ H , − ) to (26). Because I H is simple and is not punctual, Hom ( R Σ H , I H ) = 0 . The resulting exact sequence begins
0 ⟶ Hom ( R Σ H , P H ) ⟶ End ( R Σ H ) → ∂ ϵ Ext 1 ( R Σ H , I H ) , ∂ ϵ ( ϕ ) = ϵ Σ ∘ ϕ .
All Hom and Ext groups here are taken in MHM ( X 0 ) . Since the points are disjoint and End ( Q H ( − 2 ) ) = Q , we have End ( R Σ H ) = ∏ p ∈ Σ Q , and additivity of Ext 1 identifies ∂ ϵ ( ( a p ) p ) = ( a p ϵ p ) p . Consequently,
Hom ( R Σ H , P H ) ≃ ⨁ ϵ p = 0 Q .
In particular, if every ϵ p is nonzero, then Hom ( R Σ H , P H ) = 0 .
For any f ∈ End ( P H ) , equation (29) gives a unique a ∈ Q with f u = u a . Consequently ( f − a id P H ) u = 0 . Since v : P H ↠ R Σ H is the cokernel of u, f − a id P H factors through a morphism R Σ H → P H , which must be zero. Hence f = a id P H , proving (28). The only idempotents in Q are 0 and 1, so P H is indecomposable.
Conversely, if ϵ p = 0 for some p, the calculation (31) already produces a nonzero map R p H → P H whose composite with the quotient v is the inclusion R p H ↪ R Σ H . This exhibits a splitting of the R p H -component of the extension. Equivalently, by additivity of Ext 1 in its first variable, the class has components ( 0 , ϵ ′ ) under R Σ H = R p H ⊕ R Σ ∖ { p } H ; hence P H ≃ R p H ⊕ P ′ H , where P ′ H contains I H as a nonzero subobject. Both summands are nonzero, so P H is decomposable. If Σ is empty, the criterion is vacuous and P H = I H is simple. □
Remark 12
(Different punctual normalizations). The punctual forward quotient Q H ( − 2 ) in (26) is distinct from the Q H ( − 1 ) contribution occurring in the perverse-normalized nearby–vanishing-cycle interface. An isomorphism of underlying rational vector spaces does not identify these objects in MHM ( { p } ) , and the difference cannot be suppressed in a Gysin or costalk computation [2,6].
The geometric and categorical provenance used below is as follows.
  • In [3], the projective minuscule G L 2 Schubert-convolution calculation takes place in derived pure Hodge modules on a projective Schubert target. Its exceptional punctual contribution has Tate twist ( − 1 ) . The separate relative affine G L 2 collision is treated in derived mixed Hodge modules on its incidence and contraction spaces. The G L 3 tower X 2 → r X 1 → q X 0 produces coefficient-comparison arrows and an octahedral composition defect in D b MHM ( X 0 ) . The vanishing of that defect in a chosen local model does not imply splitting of its individual coefficient extensions.
  • In [4], the universal determinant hypersurface
    X det = { A ∈ Mat 2 ( C ) : det A = 0 }
    is compared with the invariant-line incidence variety
    I = { ( A , ℓ ) ∈ Mat 2 ( C ) × P 1 : A ℓ ⊆ ℓ } .
    A framed Higgs evaluation map and its incidence pullback furnish a one-step Hecke correspondence on specified algebraic charts. The auxiliary kernel-line projection q line has distinct *- and !-images of the attaching arrow. Localized nonvanishing on the genuine modified-Higgs target and the higher perverse-degree attachment on the smaller admissible open are separate statements, subject to the descent and support hypotheses in that paper. Neither assertion is a statement about an arbitrary unframed quotient stack.
  • In [5], the determinantal matrix-factorization model belongs to a two-periodic dg category; the Fourier–Laplace and Stokes constructions belong to irregular connection and Stokes-theoretic settings; and the minuscule G L 2 and stabilized A 2 constructions have their respective incidence and critical-value geometries. Only the explicitly constructed geometric maps and coefficient comparisons from that work may be used below. In particular, matching spectral values or exceptional P 1 fibers does not yield a functor into D b MHM ( X 0 ) .
Remark 13
(No implicit identification of bases). The symbols X 0 , X det , I , the framed source and target Higgs spaces, and the Schubert convolution targets denote different algebraic spaces. A pullback or direct image between their coefficient categories is used only after its defining geometric map has been specified. In particular, no arrow D b MHM ( X 0 ) → D b MHM ( M − ) is inferred from the existence of a Hecke correspondence on M − .

3.2. Canonical Direct-Image Extension and Its Actual Boundary

Let i : Σ ↪ X 0 be the reduced closed immersion and j : U ↪ X 0 its complementary open immersion. For the intersection Hodge module I H , set
A I : = i * i ! I H , B I : = j * j * I H , S Σ H : = ⨁ p ∈ Σ i p * Q H ( − 3 ) .
Here A I , B I belong to D b MHM ( X 0 ) , whereas S Σ H and R Σ H belong to the heart. The notation H q in this subsection denotes cohomology for the standard t-structure on D b MHM ( X 0 ) , whose heart is MHM ( X 0 ) ; its rational realization is perverse cohomology.
Lemma 2
(Degreewise recollement calculation). Assume the ordinary-double-point costalk computation of Appendix A, with its stated orientation and Tate conventions. Then
H q ( A I ) = R Σ H , q = 1 , S Σ H , q = 3 , 0 , q ≠ 1 , 3 .
The canonical counit and unit of the recollement adjunctions give
A I → counit I H → unit B I → ∂ A I [ 1 ] .
The nonzero cohomology objects of B I are
H q ( B I ) = P can H , q = 0 , S Σ H , q = 2 , 0 , q ≠ 0 , 2 , P can H : = H 0 ( B I ) ,
and the degree-zero part of the long exact sequence is the short exact sequence
0 ⟶ I H → u can P can H → v can R Σ H ⟶ 0 .
Proof. 
The immersion i has finite zero-dimensional source, so i * is t-exact and preserves the calculated costalk degrees. Thus H q ( A I ) = ⨁ p i p * H q ( i p ! I H ) , which gives (35). The maps in (36) are, respectively, the counit of ( i * , i ! ) and the unit of ( j * , j * ) ; the localization triangle is their distinguished cofiber sequence [6,10,18]. Since I H lies in the heart, the long exact sequence has the following nontrivial portions (the zeros are included):
0 = H 0 ( A I ) ⟶ I H → u can H 0 ( B I ) → v can R Σ H ⟶ H 1 ( I H ) = 0 , R Σ H ⟶ 0 ⟶ H 1 ( B I ) ⟶ 0 , 0 = H 2 ( A I ) ⟶ 0 ⟶ H 2 ( B I ) → ∼ S Σ H ⟶ 0 .
In degrees below zero and above two, the same sequence gives H q ( B I ) = 0 . Exactness of the first row gives (38), where u can is induced by the unit, and v can by the connecting homomorphism H 0 ( B I ) → H 1 ( A I ) . The third row gives the S Σ H constituent in degree two. No splitting of B I in the derived category is asserted. □
Proposition 9
(Construction of the canonical attaching arrow). Let τ ≤ 1 denote the standard truncation in D b MHM ( X 0 ) . There is a canonical identification τ ≤ 1 A I ≃ R Σ H [ − 1 ] and a composite
β can : R Σ H [ − 1 ] ≃ τ ≤ 1 A I ⟶ A I → counit I H .
With the cone convention of Definition 3 and the standard sign in rotation of a distinguished triangle, the Yoneda connecting morphism of (38) is
ϵ can = − β can [ 1 ] : R Σ H ⟶ I H [ 1 ] .
For each node, derived adjunction ( i p * , i p ! ) carries ϵ can , p to the negative of the canonical truncation inclusion on H 1 ( i p ! I H ) :
H 0 ( i p ! ϵ can , p ) = − id Q H ( − 2 )
after the natural identification H 1 ( i p ! I H ) ≃ Q H ( − 2 ) . In particular ϵ can , p ≠ 0 for every p.
Proof. 
By Lemma 2, the only cohomology object of τ ≤ 1 A I is R Σ H in degree one. Its canonical truncation map to A I is therefore (39) after the indicated shift. Write C β : = Cone ( β can ) . For the composable morphisms R Σ H [ − 1 ] → A I → I H , the octahedral axiom, with the chosen mapping-cone signs, gives a diagram of triangles whose relevant third triangle is
S Σ H [ − 3 ] ⟶ C β ⟶ B I ⟶ S Σ H [ − 2 ] .
Indeed, write f : R Σ H [ − 1 ] → A I for the truncation map and g : A I → I H for the counit; then g f = β can . Their four octahedral triangles are
R Σ H [ − 1 ] → f A I ⟶ S Σ H [ − 3 ] ⟶ R Σ H , A I → g I H ⟶ B I ⟶ A I [ 1 ] , R Σ H [ − 1 ] → g f I H ⟶ C β ⟶ R Σ H , S Σ H [ − 3 ] ⟶ C β ⟶ B I ⟶ S Σ H [ − 2 ] .
To identify the third triangle’s comparison maps explicitly, choose cochain representatives for f and g. With Cone ( f ) n = A I n ⊕ R Σ H [ − 1 ] n + 1 , the natural cone maps are
k : Cone ( f ) ⟶ Cone ( g f ) , k ( a , r ) = ( g ( a ) , r ) , ℓ : Cone ( g f ) ⟶ Cone ( g ) , ℓ ( b , r ) = ( b , f ( r ) ) .
They commute with the cone differentials because f and g are chain maps. Their composite has degree-minus-one homotopy h ( a , r ) = ( 0 , a ) as a map from Cone ( f ) to Cone ( g ) :
d Cone ( g ) h + h d Cone ( f ) = ℓ k .
Indeed the two components of the left side are g ( a ) and f ( r ) , respectively. These maps and the homotopy give the identified octahedral triangle, with the last rotation carrying the standard sign. Its long exact cohomology sequence shows that C β lies in the heart and fits into an exact sequence with endpoints I H and R Σ H . From (42), the map H 0 ( C β ) → H 0 ( B I ) is an isomorphism; the octahedral structure maps identify its endpoint maps with u can and v can . Rotating the cone triangle R Σ H [ − 1 ] → β can I H → C β → R Σ H to the right gives the connecting arrow − β can [ 1 ] , proving (40). For a fixed p, adjunction sends the composite R Σ H [ − 1 ] → i * i ! I H → I H to the truncation Q H ( − 2 ) [ − 1 ] → i p ! I H ; on degree one it is the identity. The rotation sign gives (41). The latter is an isomorphism and hence is nonzero. □
Remark 14
(The sign is part of the normalization). The minus sign in (40) follows from rotating the cone triangle with the shift convention K [ 1 ] n = K n + 1 . Replacing that convention or replacing the chosen identification of H 1 ( i p ! I H ) with its Tate model changes the coordinate description. Equality with a second marked extension must be checked using thesameconventions, not by renormalizing its quotient after the comparison.
Lemma 3
(The canonical middle term has no punctual subobject). The canonical truncation triangle and recollement imply
i ! P can H ≃ S Σ H [ − 3 ] .
In particular, for every p ∈ Σ and every Hodge structure L ∈ MHM ( { p } ) ,
Hom MHM ( X 0 ) i p * L , P can H = 0 .
Proof. 
Lemma 2 gives the canonical truncation triangle
P can H ⟶ B I ⟶ S Σ H [ − 2 ] ⟶ P can H [ 1 ] .
By recollement i ! B I = i ! j * j * I H = 0 . Applying i ! gives a triangle i ! P can H → 0 → S Σ H [ − 2 ] → i ! P can H [ 1 ] , so its last arrow is an isomorphism and i ! P can H ≃ S Σ H [ − 3 ] . The adjunction i p * ⊣ i p ! identifies a proposed punctual subobject map with a morphism from the heart object L to Q H ( − 3 ) [ − 3 ] in D b MHS Q . That Hom group vanishes by the standard t-structure. Thus no nonzero punctual morphism into P can H exists. □

3.3. MacPherson–Vilonen Gluing and Marked Extensions

The MacPherson–Vilonen construction records an open-stratum perverse object together with punctual gluing data; the appropriate recollement and link complexes must be retained for an isolated threefold singularity [18,19]. The comparison here is made first by actual adjunction maps in MHM ( X 0 ) . The corresponding rational zigzags give a local check, but are not substituted for an isomorphism of Hodge modules.
Proposition 10
(Canonical map from a corrected middle term). Let (26) be the specified corrected sequence and identify j * P H with j * I H by j * u . There is a canonically constructed morphism
a P : P H ⟶ P can H
in MHM ( X 0 ) , determined by the adjunction unit P H → j * j * P H and the standard truncation of B I . It extends the canonical inclusion of I H and induces a punctual quotient endomorphism
h R : R Σ H ⟶ R Σ H , h R = ⨁ p ∈ Σ h p , h p ∈ Q .
These maps form a diagram of short exact sequences
0 → I H → u P H → v R Σ H → 0 id ↓ id a P ↓ a P h R ↓ h R 0 → I H → u can P can H → v can R Σ H → 0 .
Its connecting maps satisfy
ϵ can ∘ h R = ϵ Σ .
The morphism a P is an isomorphism if and only if all h p are nonzero. It is an isomorphism of thefixed marked extensions(identity on both endpoints) precisely when h R = id , equivalently ϵ Σ = ϵ can .
Proof. 
The unit of ( j * , j * ) gives a morphism P H → j * j * P H ≃ B I . By Lemma 2, B I ∈ D ≥ 0 and its truncation τ ≤ 0 B I = H 0 ( B I ) = P can H . For a heart object P H ∈ D ≤ 0 , the canonical truncation adjunction gives
Hom ( P H , P can H ) → ∼ Hom ( P H , B I ) .
The image of the unit under the inverse of this isomorphism is a P . Naturality of the unit in the morphism u : I H → P H shows a P u = u can . Since v can a P u = 0 , the universal property of the cokernel v : P H ↠ R Σ H in the abelian category MHM ( X 0 ) gives a unique h R such that v can a P = h R v . By support on the pairwise disjoint closed points, the endomorphism algebra of R Σ H is the product of the scalar endomorphism algebras of its summands, so h R = ⊕ h p . This proves the two left squares and the quotient square of (48). Functoriality of the Yoneda connecting arrow for a morphism of short exact sequences gives the commutative last square, which is (49). The first and third vertical maps in the diagram are isomorphisms exactly when h R is invertible. The short five lemma in the abelian category then shows a P is an isomorphism. Conversely, if a P is an isomorphism and its restriction to I H is the identity, its induced quotient map is an isomorphism. Finally, agreement as marked extensions fixes the quotient map to be the identity. By the nonzero coordinate of ϵ can at every node, the equality of connecting maps is equivalent to h R = id . □
Corollary 1
(Local criterion in the normalized costalk). For every p, let c p ∈ Q be defined by the actual costalk map
H 0 ( i p ! ϵ p ) = c p id Q H ( − 2 ) .
With the sign convention of Proposition 9,
h p = − c p .
Consequently P H ≃ P can H via the canonical morphism a P if and only if c p ≠ 0 for every node, whereas the identity-marked equality holds if and only if c p = − 1 for all p.
Proof. 
Apply the adjunction i p * ⊣ i p ! to (49) and take H 0 after shifting the target by one. The canonical map is − id by (41); hence ( − id ) h p = c p id . The conclusions follow from Proposition 10. □
Lemma 4
(Costalk and stalk tests for the gluing class). At an ordinary double point, the corrected extension yields
H 0 ( i p ! P H ) ≃ ker ( c p : Q H ( − 2 ) → Q H ( − 2 ) ) , H 1 ( i p ! P H ) ≃ coker ( c p : Q H ( − 2 ) → Q H ( − 2 ) ) , H 3 ( i p ! P H ) ≃ Q H ( − 3 ) ,
with the other costalk cohomology objects zero. Its stalk has
H − 3 ( i p * P H ) ≃ Q H , H − 1 ( i p * P H ) ≃ Q H ( − 1 ) , H 0 ( i p * P H ) ≃ Q H ( − 2 ) ,
and no other nonzero cohomology objects. For the canonical extension c p = − 1 , so both groups in degrees zero and one in (53) vanish.
Proof. 
Apply i p ! to the corrected distinguished triangle. The entire part of its long exact sequence in degrees zero and one is
0 ⟶ H 0 ( i p ! P H ) ⟶ Q H ( − 2 ) → c p Q H ( − 2 ) ⟶ H 1 ( i p ! P H ) ⟶ 0 .
The next relevant portion is 0 = H 2 ( i p ! R Σ H ) → H 3 ( i p ! I H ) → H 3 ( i p ! P H ) → 0 ; all remaining degrees vanish by Appendix A. Applying i p * instead gives, in degrees − 3 , − 1 , 0 ,
H − 3 ( i p * I H ) → ∼ H − 3 ( i p * P H ) , H − 1 ( i p * I H ) → ∼ H − 1 ( i p * P H ) , 0 ⟶ H 0 ( i p * P H ) → ∼ Q H ( − 2 ) ⟶ 0 .
These identities and the ODP stalk table give (54). □
Remark 15
(The precise MacPherson–Vilonen comparison). The recollement triangle for any such P H also yields the local distinguished triangle
i p ! P H ⟶ i p * P H ⟶ i p * j * j * P H ⟶ i p ! P H [ 1 ] .
Its long exact sequence includes the gluing maps
H − 1 ( i p * j * j * P H ) ⟶ H 0 ( i p ! P H ) ⟶ H 0 ( i p * P H ) ⟶ H 0 ( i p * j * j * P H ) ⟶ H 1 ( i p ! P H ) .
The link groups in this display have Tate types Q H ( − 1 ) and Q H ( − 2 ) in the respective degrees. The maps, rather than just these groups, distinguish the local extension. The rationalized diagram is the local information used by the MacPherson–Vilonen gluing formalism. It is not by itself a claim of full faithfulness on all Hodge-module morphisms. The stronger marked comparison in Proposition 10 is constructed inside MHM ( X 0 ) ; its remaining input is the actually calculated costalk scalar c p of the corrected variation-cone extension.

3.4. Selection of the Actual MHM Transport, before Any New Source Category

Definition 16
(Admissible framed transport datum). Fix finite-type complex algebraic varieties M − , H μ , and M + with morphisms
M − ← p − H μ → q + M + .
Let a − : M − → X det be the algebraic evaluation morphism on a chosen framed chart, and let ϵ det : R det H → I det H [ 1 ] be the corrected determinant-conifold arrow in D b MHM ( X det ) . Fix an object K μ ∈ D b MHM ( H μ ) and a prescribed cohomological shift and Tate twist, incorporated in K μ . Set
I − H = a − * I det H , P − H = a − * P det H , R − H = a − * R det H , ϵ − = a − * ( ϵ det ) : R − H ⟶ I − H [ 1 ] .
These are objects and an arrow in D b MHM ( M − ) . No perverse-heart assertion is made for the pullbacks without the requisite t-exactness hypotheses.
For Z = M − , H μ , M + write D H ( Z ) for the bounded stable enhancement of D b MHM ( Z ) furnished by the compatible six-functor framework. Saito’s operations apply to the displayed finite-type algebraic maps, and Tubach provides their stable enhancements and coherent exchange transformations [6,10]. Thus the star-normalized operator
H μ , * ( M ) = q + * K μ ⊗ p − * M , M ∈ D H ( M − ) ,
is an exact functor to D H ( M + ) whenever the selected bounded subcategories are preserved. Here ⊗ denotes the derived MHM tensor product. If exceptional pullback and proper-support pushforward are required, a separately normalized operator H μ , ! is defined using the specified ! operations; the two operators are not conflated.
Proposition 11
(Exact transport of the designated arrow). Assume the datum of Definition 16 with the boundedness condition above. Then
H μ , * ( I − H ) ⟶ H μ , * ( P − H ) ⟶ H μ , * ( R − H ) → H μ , * ( ϵ − ) H μ , * ( I − H ) [ 1 ]
is a distinguished triangle in D b MHM ( M + ) . The associated attaching morphism is the image of thespecifiedarrow ϵ − , not an independently selected element of the target Hom group.
Proof. 
The exact sequence on X det yields a fiber sequence in D H ( X det ) . The exact functor a − * preserves finite fibers, suspensions, and cofibers. It therefore carries that sequence to the triangle of (59), with connecting arrow a − * ϵ det . The functors p − * and q + * are exact on their stable enhancements; tensor product with the fixed kernel K μ is exact as well. Their composite therefore carries this triangle to a fiber/cofiber sequence. Compatibility with suspension identifies its last arrow with H μ , * ( ϵ − ) . Passing to homotopy categories gives (61). □
Proposition 12
(A local smooth-pullback detection test). Let e : S → X det be a smooth morphism of pure relative dimension d on an algebraic chart, and let s ∈ S map to the node o ∈ X det . Write e † = e * [ d ] for the perverse-normalized smooth pullback. If the local costalk morphism
H 0 ( i o ! ϵ det ) : Q H ( − 2 ) ⟶ H 1 ( i o ! I det H )
is nonzero, then e † ( ϵ det ) is nonzero in D b MHM ( S ) . This applies in particular to the canonical extension of Proposition 9.
Proof. 
Smooth morphisms are, locally in the smooth/étale topology, projections from a product with a smooth relative d-dimensional factor. Pass to such a neighborhood of s and use smooth base change for the supported Hodge modules; the relative-dimension shift and the associated Tate conventions are those of the normalized pullback. The pullback of the local attachment is the external product of its costalk morphism with the perverse constant Hodge module Q A d H [ d ] on the smooth factor. Apply the costalk at s. The factor in the relative direction is
i s ! Q A d H [ d ] ≃ Q H ( − d ) [ − d ] ,
so the map obtained from (62) is the nonzero original map tensored with the one-dimensional Tate structure Q H ( − d ) , with its common cohomological shift. Tensoring a nonzero rational linear map with a one-dimensional nonzero rational vector space is nonzero. If e † ( ϵ det ) were zero, every such costalk image would vanish, a contradiction. The canonical case uses (41). □
Remark 16
(Heart versus derived attachment). The terms of (61) need not belong to the perverse heart. Consequently H μ , * ( ϵ − ) need not represent an element of Ext MHM ( M + ) 1 between its displayed endpoints. A higher Postnikov connecting morphism obtained from perverse truncation is a further derived invariant, not a change in the definition of the source Yoneda class.

3.5. Construction by Gluing and Correspondences

Let U = X 0 ∖ Σ and write j : U ↪ X 0 , i : Σ ↪ X 0 . For M ∈ D H ( X 0 ) the localization identities give two canonical distinguished triangles
j ! j * M ⟶ M ⟶ i * i * M → ∂ M * j ! j * M [ 1 ] ,
i * i ! M ⟶ M ⟶ j * j * M → ∂ M ! i * i ! M [ 1 ] .
The identities j ! ≃ j * and i ! ≃ i * are used here; they do not identify j ! with j * or i * with i ! . The recollement transformations satisfy
j * i * = 0 , i * j ! = 0 , i ! j * = 0 , j * j ! ≃ id , j * j * ≃ id , i * i * ≃ i ! i * ≃ id .
These are the established localization operations in the derived MHM category [6,10,18].
Lemma 5
(Open and punctual restrictions of the corrected extension). For the triangle (25),
j * I H → j * u j * P H ⟶ 0 ⟶ j * I H [ 1 ]
is distinguished; in particular j * u is an isomorphism. For each p ∈ Σ there are distinguished triangles
i p * I H ⟶ i p * P H ⟶ Q H ( − 2 ) → i p * ϵ Σ i p * I H [ 1 ] ,
i p ! I H ⟶ i p ! P H ⟶ Q H ( − 2 ) → i p ! ϵ Σ i p ! I H [ 1 ] .
Proof. 
Because R Σ H is supported on Σ , j * R Σ H = 0 . Apply the exact functor j * to (25); the first two terms are therefore isomorphic. Next, apply the exact functors i p * and i p ! to the same triangle. For q ≠ p , disjoint support gives i p * i q * = i p ! i q * = 0 . Kashiwara’s equivalence and closed immersion adjunction give i p * i p * ≃ i p ! i p * ≃ id . Thus both images of R Σ H are Q H ( − 2 ) , and their connecting morphisms are precisely the corresponding images of ϵ Σ . □
A comparison with a variation-cone construction requires more than (65). Let V denote the normalized variation-cone object of Section 2.5, together with the comparison maps provided by [2]. To identify its middle term with P H , one must exhibit morphisms f I , f P , f R making a map of triangles
I V → P V → R V → ϵ V I V [ 1 ] f I ↓ f I f P ↓ f P f R ↓ f R f I [ 1 ] ↓ f I [ 1 ] I H → u P H → v R Σ H → ϵ Σ I H [ 1 ] .
Here the top row is the distinguished triangle produced by the specified normalized variation and truncation construction; no arbitrary cone of variation is asserted to lie in MHM ( X 0 ) . An identification requires f I , f R to be isomorphisms and the commutativity of the final square. The comparison maps and their normalizations are inputs to be verified against the corrected source construction; they are not created by choosing a scalar after the fact.
Remark 17
(Conditional comparison with an independent category). Only if an independently geometric stable category C att has been defined, with a morphism α : C → A [ 1 ] , do we consider an exact functor
R H : C att ⟶ D b MHM ( X 0 ) .
The source must have its own objects, mapping data, and operations; a thick arrow subcategory generated by ϵ Σ inside D b MHM ( X 0 ) does not constitute an independent comparison. This optional functor is not required to establish Proposition 11.

3.6. The Comparison of Actual Attaching Morphisms

Definition 17
(Soundness of an attachment comparison). Assume the independent category, objects, and functor of Remark 17. An attachment comparison consists of isomorphisms
η A : R H ( A ) → ∼ I H , η C : R H ( C ) → ∼ R Σ H ,
for which the diagram
R H ( C ) → R H ( α ) R H ( A ) [ 1 ] η C ↓ η C η A [ 1 ] ↓ η A [ 1 ] R Σ H → ϵ Σ I H [ 1 ]
commutes in D b MHM ( X 0 ) , with a coherent square in the chosen enhancements when composition requires it. We call this conditionsoundness.Coverageis the separate assertion that the independently defined source contains the specified geometric attachments and is stable under the selected operations.
Proposition 13
(Attachment comparison from independently normalized triangles). Suppose there is a distinguished triangle in C att ,
A ⟶ B ⟶ C → α A [ 1 ] ,
and a morphism of its image under R H to (25) with endpoint components η A , η C from (71). Then (72) commutes. If the endpoint components are isomorphisms, the middle component is an isomorphism.
Proof. 
A morphism of distinguished triangles is a family of morphisms commuting with each of the four structure arrows, including the connecting arrows. Its last square is precisely (72). If the first and third vertical maps are isomorphisms, apply the cohomological functor Hom ( T , − ) , for an arbitrary object T, to both triangles. The resulting long exact sequences form a commutative diagram. The five lemma, applied to consecutive five-term portions, shows that Hom ( T , − ) takes the middle vertical map to an isomorphism in every degree. Yoneda’s lemma implies that the middle vertical map is an isomorphism. □
Remark 18
(Normalization and scalar ambiguity). If the relevant Hom group happens to be one-dimensional, two nonzero arrows differ by a scalar only after their endpoints have been identified. That observation does not establish (72). The endpoint maps, orientation or trace normalization, and the resulting scalar must be obtained from the geometry and checked; they are not chosen to force the required equality.

3.7. One-node Ext Calculation and Normalization

Fix p ∈ Σ and write i = i p . The local attachment group is
E p : = Hom D b MHM ( X 0 ) i * Q H ( − 2 ) , I H [ 1 ] .
Proposition 14
(Costalk reduction). There is a canonical isomorphism
E p ≃ Hom D b MHM ( { p } ) Q H ( − 2 ) , i ! I H [ 1 ] .
Under this isomorphism, the component ϵ p of ϵ Σ corresponds to i ! ( ϵ p ) composed with the counit identification i ! i * ≃ id .
Proof. 
For the closed immersion i, the derived adjunction ( i * , i ! ) gives, naturally in L ∈ D b MHM ( { p } ) and M ∈ D b MHM ( X 0 ) ,
Hom D b MHM ( X 0 ) ( i * L , M ) ≃ Hom D b MHM ( { p } ) ( L , i ! M ) .
Take L = Q H ( − 2 ) and M = I H [ 1 ] . Naturality of adjunction identifies the image of a morphism i * L → I H [ 1 ] with its costalk image, after the counit isomorphism. This proves both assertions. □
Corollary 2
(The ODP extension line and its canonical generator). For the prescribed punctual Tate coefficient at an ordinary double point, there are canonical isomorphisms
E p ≃ Hom D b MHS Q Q H ( − 2 ) , i p ! I H [ 1 ] ≃ Hom MHS Q Q H ( − 2 ) , Q H ( − 2 ) ≃ Q .
The canonical attaching morphism of Proposition 9 is a nonzero generator; in the costalk convention fixed there its coordinate is − 1 . This calculation is specific to the ordinary double point and the fixed coefficient Q H ( − 2 ) .
Proof. 
The first map is the adjunction isomorphism of Proposition 14. Appendix A gives H 1 ( i p ! I H ) = Q H ( − 2 ) , H 3 ( i p ! I H ) = Q H ( − 3 ) , and all other cohomology zero. After shift by [ 1 ] , the two cohomology degrees are 0 and 2. Let K = i p ! I H [ 1 ] . The truncation triangle τ ≤ 0 K → K → τ ≥ 1 K and the orthogonality of the standard t-structure give Hom ( Q H ( − 2 ) , K ) ≃ Hom ( Q H ( − 2 ) , H 0 K ) . Here H 0 K = Q H ( − 2 ) , whose rational Hodge-structure endomorphisms form Q . Formula (41) identifies the canonical morphism with the negative identity in this group. □
Remark 19
(What the dimension does not determine). The dimension of E p determines neither the coordinate c p of the corrected variation-cone class nor the marked comparison with the canonical exceptional-curve generator. The geometric Gysin comparison itself is proved under the projective small-resolution hypothesis in Theorem 1. The still-unfixed corrected coordinate is computed from the actual costalk morphism of ϵ p ; see (51).
Definition 18
(Normalized local comparison). The canonical extension supplies the specified nonzero arrow g p : = ϵ can , p : i * Q H ( − 2 ) → I H [ 1 ] . A normalized local comparison is the equality
ϵ p = λ p g p , λ p ∈ Q ,
where the orientation, boundary convention, source Tate twist, and identification of the costalks are specified independently. By Corollary 2, g p is a generator and the scalar is uniquely determined. The relation with the actual costalk scalar is λ p = − c p . When a projective small resolution with smooth source exists, the exceptional-curve Gysin arrow of Proposition A9 has costalk coordinate + 1 and therefore equals − g p with these conventions. This comparison does not determine the corrected marked scalar λ p .
Proposition 15
(Unfixed marked scalar and invariance of local nonsplitting). Assume that the specified corrected local class ϵ p is nonzero. Then, with the independently fixed costalk and cone conventions of Proposition 9, there is a unique λ p ∈ Q × such that
ϵ p = λ p ϵ can , p , c p = − λ p , h p = λ p .
The canonical comparison of Proposition 10 is an isomorphism of middle Hodge modules at p, but it preserves both specified endpoint markings if and only if λ p = 1 . The local filtered and associated-graded nonsplitting results of Theorem 3, below, do not depend on this last equality.
Proof. 
Corollary 2 identifies the local extension group with Q , and the canonical arrow has coordinate − 1 . The actual corrected costalk coordinate is c p ; therefore λ p = − c p , while Corollary 1 gives h p = − c p . Nonvanishing of ϵ p is equivalent to λ p ≠ 0 . The five-lemma and marking assertions follow from Proposition 10. For the last assertion, an invertible rational automorphism of the punctual quotient transports the canonical strict filtered exact sequence to the corrected one. Such a change of endpoint marking cannot change whether the associated-graded extension is split. The explicit graded nonsplitting is proved in Theorem 3; no numerical value of λ p is used here. □
Remark 20
(What remains of the marked variation-cone comparison). The equality λ p = 1 is not a consequence of the one-dimensional extension group or of the nonvanishing of the corrected class. It requires the comparison of the actual corrected variation-cone boundary with the independently normalized recollement counit, including the can / var maps, their cone rotations, and the Tate-twisted costalk coordinate. The exceptional-curve Gysin normalization of Appendix A.8 fixes the canonical geometric generator, but does not supply the missing comparison with the corrected variation-cone boundary. We retain the scalar as part of the marked data rather than redefining the quotient generator to force equality.

3.8. Restricted Transport Proposition and Exactness Ledger

Proposition 16
(Restricted one-step statement). Let the framed correspondence and kernel be as in Definition 16. Let k : V ↪ M + be an algebraic open immersion. The transported arrow on V is
ϵ μ , V : = k * H μ , * ( ϵ − ) : k * H μ , * ( R − H ) ⟶ k * H μ , * ( I − H ) [ 1 ] .
Its defining triangle is the restriction of (61). If the localized nonvanishing theorem of [4] applies to this precise choice of geometric charts, coefficient kernel, normalization, and target open V, then ϵ μ , V ≠ 0 in D b MHM ( V ) .
Proof. 
Apply the exact functor k * to (61); its last arrow is (79). The final assertion is the specified localized nonvanishing result from [4], not a consequence of exactness. □
Remark 21
(Exactness and applicability). The following assertions must be checked separately:
  • Algebraic applicability:the maps defining (58) belong to the finite-type algebraic six-functor setting. A statement on an unframed quotient stack requires its own equivariant or stack-theoretic justification.
  • Categorical exactness:pullback, tensor with the fixed kernel, pushforward, and restriction are exact on the selected stable categories. This is independent of perverse t-exactness.
  • Soundness of the arrow:the transported morphism is the image of ϵ − under these operations, with the original normalizations. Its nonvanishing is a further geometric assertion.
  • Coverage:extending the construction from the chosen chart and open V to additional Higgs spaces, nodes, or repeated modifications requires independent geometric verification.

3.9. Global Node Relations and Morphism-Level Transport

Let a : X 0 → Spec C be the proper structure morphism. The derived pushforward a * : D b MHM ( X 0 ) → D b MHS Q is defined by Saito’s proper direct image. It carries the corrected attaching morphism to
a * ( ϵ Σ ) : a * R Σ H ⟶ ( a * I H ) [ 1 ] .
Since the punctual sources are point Hodge modules, H 0 ( a * R Σ H ) = ⨁ p Q H ( − 2 ) . With I H normalized by the threefold shift, set
V Σ H : = ⨁ p ∈ Σ Q H ( − 2 ) , I H 4 ( X 0 , Q ) : = H 1 ( a * I H ) , ∂ Σ H : = H 0 ( a * ϵ Σ ) : V Σ H ⟶ I H 4 ( X 0 , Q ) ,
Define the corresponding relation kernel by
L Σ H : = ker ∂ Σ H .
The target is a rational mixed Hodge structure; under properness and the intersection-complex purity hypotheses it is pure of weight four. This map is induced by an actual connecting morphism, not by choosing a linear relation space after the calculation. In particular, a tuple ( a p ) p ∈ Σ represents a relation precisely when
∑ p ∈ Σ a p ∂ Σ H ( e p ) = 0 ,
where e p is the standard Tate-basis vector in V Σ H after the fixed punctual identifications.
Proposition 17
(Construction of the global relation exact sequence). The global sections of (25) produce the exact sequence of mixed Hodge structures
H 0 ( a * I H ) ⟶ H 0 ( a * P H ) ⟶ V Σ H → ∂ Σ H I H 4 ( X 0 , Q ) ⟶ H 1 ( a * P H ) ⟶ H 1 ( a * R Σ H ) = 0 .
In particular,
0 ⟶ L Σ H ⟶ V Σ H → ∂ Σ H im ∂ Σ H ⟶ 0
is short exact in MHS Q .
Proof. 
The derived functor a * is exact on its stable enhancement and therefore takes the source triangle to a distinguished triangle. Taking the standard cohomology long exact sequence in D b MHS Q yields (84); the rightmost zero follows because a * R Σ H = V Σ H is concentrated in degree zero. The definition of kernel and image in the abelian category of mixed Hodge structures gives the short exact sequence (85). □
Proposition 18
(Diagonal rescaling of the rational relation map). Let ϵ Σ , can be the direct sum of the canonical local attaching arrows with the fixed punctual markings, and suppose all components of the corrected attachment are nonzero. Write ϵ p = λ p ϵ can , p as in (78) and define the rational Hodge-structure automorphism
D λ : = ⨁ p ∈ Σ λ p id Q H ( − 2 ) : V Σ H → ∼ V Σ H .
Then the connecting maps induced by proper derived global sections obey
∂ Σ , corr H = ∂ Σ , can H ∘ D λ , L Σ , corr H = D λ − 1 L Σ , can H .
In particular the ranks of the rational connecting maps and the rational dimensions of their kernels agree. The embedded coordinate subspaces need not be equal. This assertion has no implication for the saturation or torsion of integral relation lattices.
Proof. 
Because R Σ H is a direct sum of punctual Hodge modules, the specified arrow is the sum of its restrictions to the summands. Consequently ϵ Σ , corr = ϵ Σ , can ∘ D λ in D b MHM ( X 0 ) , identifying the direct-sum punctual source with V Σ H after proper pushforward. The exact derived functor a * preserves composition, and standard cohomology sends this identity to the first formula in (87). Since every λ p ≠ 0 , D λ is invertible over Q ; taking kernels gives the second formula. Rank and dimension are invariant under precomposition by an automorphism, whereas equality of the actual embedded kernels requires further information. No integral version of D λ has been constructed. □
Proposition 19
(Relation compatibility under a specified morphism of triangles). Suppose X 0 ′ is a second projective nodal threefold with a specified MHM attachment, and suppose independently constructed geometric operations give morphisms on the corresponding global connecting-map targets
V Σ H → ∂ Σ H I H 4 ( X 0 , Q ) u V ↓ u V u I ↓ u I V Σ ′ H → ∂ Σ ′ H I H 4 ( X 0 ′ , Q ) .
If these two maps arise from a morphism of the correspondingderived global attachment triangles, the square commutes and u V ( L Σ H ) ⊆ L Σ ′ H .
Proof. 
A morphism of distinguished triangles commutes with their connecting arrows. Applying H 0 to the resulting global morphism gives u I ∂ Σ H = ∂ Σ ′ H u V . For v ∈ L Σ H the right side vanishes, hence ∂ Σ ′ H ( u V v ) = u I ∂ Σ H ( v ) = 0 . This proves the claimed kernel inclusion. Construction of the maps u V , u I in a selected bivariant geometry is a separate geometric task; the proposition does not postulate maps between arbitrary nodal families and Higgs targets. □
Theorem 1
(Exceptional-curve realization of the global relation boundary). Suppose that the projective nodal threefold X 0 admits a projective small resolution ρ : X ˜ 0 → X 0 with smooth source and exceptional curves C p = ρ − 1 ( p ) ≃ P 1 for all p ∈ Σ . Normalize the small-map identification θ ρ : I H 4 ( X 0 , Q ) → ∼ H 4 ( X ˜ 0 , Q ) by the identity on U = X 0 ∖ Σ , and use complex orientation for the codimension-two purity maps. Let γ p = PD [ C p ] and define the morphism of pure rational Hodge structures
G ρ : V Σ H ⟶ H 4 ( X ˜ 0 , Q ) , e p ⟼ γ p .
With the canonical attaching morphism of Proposition 9, one has
θ ρ ∂ Σ , can H = − G ρ , θ ρ ∂ Σ , corr H = − G ρ ∘ D λ , L Σ , corr H = D λ − 1 ker G ρ .
The minus sign is the rotation sign of (40), not a sign assigned by the curve orientation. These statements are rational and do not identify integral relation lattices, assert existence of a projective small resolution for an arbitrary nodal threefold, or determine the marked variation-cone scalars λ p .
Proof. 
Write s p : C p ↪ X ˜ 0 for the regular codimension-two embedding, b p : C p → { p } for its structure map, and i p : { p } ↪ X 0 . The square ρ s p = i p b p is Cartesian. Smallness gives a canonical identification of pure Hodge modules I H ≃ R ρ * Q X ˜ 0 H [ 3 ] , normalized over U. Absolute purity for the regular embedding gives s p ! Q X ˜ 0 H ≃ Q C p H ( − 2 ) [ − 4 ] and its counit, the Gysin morphism Gys s p : s p * Q C p H ( − 2 ) [ − 4 ] → Q X ˜ 0 H . Define the geometric arrow by the following explicit composite:
Γ p : i p * Q p H ( − 2 ) → i p * η b p i p * R b p * Q C p H ( − 2 ) ≃ R ρ * s p * Q C p H ( − 2 ) → R ρ * ( Gys s p [ 4 ] ) R ρ * Q X ˜ 0 H [ 4 ] → θ ρ − 1 [ 1 ] I H [ 1 ] .
Here η b p is the unit of ( b p * , R b p * ) , and in the last arrow θ ρ denotes the corresponding small-map identification of Hodge-module objects before global cohomology. Extraordinary proper base change, compatible with the stated purity normalization, identifies
i p ! I H [ 1 ] ≃ R b p * s p ! Q X ˜ 0 H [ 4 ] ≃ R Γ ( C p , Q C p H ( − 2 ) ) .
Under derived adjunction ( i p * , i p ! ) , the mate of Γ p is the unit Q p H ( − 2 ) → R b p * Q C p H ( − 2 ) . Its degree-zero component sends 1 to the constant section 1; therefore Γ p has costalk coordinate + 1 . By (41), the canonical arrow has coordinate − 1 in this very same one-dimensional Q -space. The isomorphism of Corollary 2 consequently gives the equality of actual derived morphisms ϵ can , p = − Γ p , with fixed endpoints.
Proper global sections of the Gysin counit send 1 to s p * ( 1 ) = PD [ C p ] in H 4 ( X ˜ 0 , Q ) , with the complex orientation and the ( − 2 ) Tate coefficient. Since the direct-sum punctual source has basis e p , the equality of the local arrows gives θ ρ ∂ Σ , can H = − G ρ column by column. Applying the diagonal-rescaling identity (87) proves the second equality. As D λ is invertible over Q , its kernel is D λ − 1 ker G ρ , establishing the third equality. The construction and its sign agree with the independently recorded purity and unit–counit calculation in Proposition A9 and Theorem A1. □
Remark 22
(Scope of the boundary–Gysin comparison). The comparison in Theorem 1 is a statement about actual morphisms: the Gysin arrow is constructed from the proper-adjunction unit and the purity counit, rather than inferred from the dimension of a local extension group. It applies only after a projective small resolution with smooth source has been specified and its Hodge-module identification normalized on U. It does not identify the corrected variation-cone arrow with the canonical arrow as amarkedextension. In particular, the unknown λ p remains in D λ , and no integral or saturation statement follows from the rational relation identity.
Remark 23
(Local Ext groups and global relations have different types). The canonical decomposition
Hom D b MHM ( X 0 ) ( R Σ H , I H [ 1 ] ) ≃ ⨁ p ∈ Σ E p
concerns the extension groups in the MHM derived category. The relation map ∂ Σ H is a morphism of global mixed Hodge structures. A nonzero tuple of local attachments may lie in the kernel of a global cohomological comparison, and the existence of a relation does not split its local components. When a projective small resolution with smooth source exists, Theorem 1 identifies this global map with the exceptional-curve Gysin map under the stated orientation and marking conventions. This identification is not available for arbitrary nodal threefolds, and it does not compute the corrected local scalars λ p .

4. Filtered D -Module Components, Rees Specialization, and Attachment Defects

The corrected MHM sequence has a specified filtered regular-holonomic D -module component on a fixed smooth ambient variety. In this section we use the exact filtered-component functor of Proposition 2, construct its Rees image, and calculate the two specialization fibers with their connecting morphisms. The Bockstein triangle gives a precise criterion for survival of the extension at z = 0 , and a Koszul resolution at the determinantal node expresses it through four multiplication maps on the actual Rees extension. We then identify the canonical middle term with a localized filtered D -module and calculate its first relevant Hodge levels. The resulting non-splitting theorem settles the local determinantal specialization for the canonical extension and for every nonzero corrected class, independently of the outstanding marked scalar. These are operations in filtered and graded module categories, not in KKPY’s category of quantum F-bundles.
The elementary tests below exhibit both possible outcomes for general strict filtered extensions; they do not replace the conifold-specific localization calculation. Our conventions use left differential modules, the shifts fixed in Section 2, and the Rees parameter z; the separate KKPY parameter u is not identified with z.

4.1. Filtered De Rham Realization

Let X = X 0 and let i : X ↪ Y be a closed algebraic embedding into a smooth complex variety of dimension d. Write O Y for the structure sheaf and D Y for the sheaf of algebraic differential operators, endowed with the increasing order filtration F p ord D Y , where F p ord D Y = 0 for p < 0 . Let Mod F good ( D Y ) denote the additive category whose objects are left D Y -modules M with an exhaustive, separated, increasing good filtration F p M by coherent O Y -modules, and whose morphisms preserve the filtration. Thus
F a ord D Y · F b M ⊆ F a + b M , gr F M = ⨁ b ∈ Z F b M / F b − 1 M
is a coherent module over gr ord D Y . We use F for the module filtration and F ord for the order filtration when both occur in one formula.
For a left regular holonomic module M, our de Rham convention is
DR Y ( M ) = M ⟶ Ω Y 1 ⊗ O Y M ⟶ … ⟶ Ω Y d ⊗ O Y M [ d ] ,
where the unshifted complex is placed in degrees 0 , … , d . For a mixed Hodge module H supported on X, let
For F , i ( H ) = ( M H , F • M H )
be its left filtered regular-holonomic D Y -module component, obtained by the exact functor For F , i : MHM ( X ) → FilRH str ( Y ) of Definition 6 and Proposition 2. On objects this is the projection of the embedded MHM datum onto ( M H , F ) ; on a morphism it is the filtered D Y -linear component. It preserves identities and compositions by the defining morphism rules, and carries MHM short exact sequences to strictly exact filtered sequences by Saito’s strictness theorem. Its rational-perverse component K H is identified with DR Y ( M H ) after complexification, with the chosen normalization. Kashiwara equivalence identifies the underlying supported differential-module objects with those on X; it does not make D X a sheaf of differential operators on a smooth variety. If the input is first described on an ambient variety Y ′ and a graph embedding γ : Y ′ ↪ Y ′ × B is used, the filtered direct image is γ + with its transfer-bimodule filtration; we use the corresponding image on the smooth ambient space and fix the resulting filtration shift before comparing morphisms [6,8,9].
Remark 24
(The information forgotten by For F , i ). The object ( M H , F ) does not retain the rational perverse sheaf, the weight filtration, or the comparison isomorphism as separate data. Consequently, the forgetful operation need not be faithful on extension classes. Its definition on the abelian category of mixed Hodge modules does not assert that arbitrary filtered complexes or arbitrary morphisms of filtered differential modules admit mixed-Hodge-module lifts.

4.2. The Exact Category in Which Filtered Extension Maps Live

Definition 19
(Strict filtered exact sequence). A sequence of good filtered left D Y -modules
0 ⟶ ( A , F ) → a ( P , F ) → b ( C , F ) ⟶ 0
isstrictly exactif the underlying sequence of D Y -modules is exact and, for every p ∈ Z ,
0 ⟶ F p A → F p a F p P → F p b F p C ⟶ 0
is exact as a sequence of O Y -modules. We use Fil Y str as a short name for the strict exact category FilRH str ( Y ) already defined in Section 2.3; its morphisms remain all filtration-preserving D Y -linear maps, not just the strict ones.
Put
( A , F ) = For F , i ( I H ) , ( P , F ) = For F , i ( P H ) , ( C , F ) = For F , i ( R Σ H ) ,
and write a and b for the filtered morphisms induced by u and v in (14).
Proposition 20
(Strictness of the corrected filtered sequence). The sequence
0 ⟶ ( A , F ) → a ( P , F ) → b ( C , F ) ⟶ 0
is a conflation of Fil Y str . Its underlying sequence of regular-holonomic D Y -modules is exact. The conclusion does not imply that the induced extension of associated graded modules is nonsplit.
Proof. 
The maps u and v are morphisms in the abelian category MHM ( X ) . Their filtered differential-module components are strict by Saito’s strictness theorem for morphisms of mixed Hodge modules. In particular,
a ( A ) ∩ F p P = a ( F p A ) , b ( F p P ) = F p C
for every p. Exactness of the underlying differential modules follows from the exactness of the mixed-Hodge-module sequence. For m ∈ F p P with b ( m ) = 0 , exactness gives m = a ( n ) for some n ∈ A ; strictness of a gives n ∈ F p A . Thus ker ( F p b ) = im ( F p a ) for every p, proving strict exactness. The statement about nonsplitting is not an exactness assertion and does not follow from strictness. □
Definition 20
(Filtered extension class). Let D b ( Fil Y str ) be the bounded derived category of the exact category Fil Y str , formed by inverting the acyclic complexes for its exact structure. The conflation (94) defines a connecting arrow
ϵ F : C ⟶ A [ 1 ] in D b ( Fil Y str ) .
It is the class of thespecifiedfiltered sequence, with the fixed endpoints and maps; it is not a choice of generator of an abstract extension space.
Remark 25
(Derived versus abelian strictness). The strictness used in Proposition 20 is a statement about each morphism in the mixed-Hodge-module heart. It does not claim that every morphism of complexes of filtered differential modules is strict in every degree, nor that an arbitrary filtered derived lift of a morphism is unique. For the particular class ϵ F , the strictly exact sequence (94) already provides the required derived representative.

4.3. Rees Algebra and the Two Fibers

Define the graded Rees algebra R Y = R F ord ( D Y ) . The central variable z has degree 1, and the Rees construction of a filtered module is
R F ( M ) = ⨁ p ∈ Z F p M z p , R F ( D Y ) = ⨁ p ≥ 0 F p ord D Y z p .
The notation z p records a grading; the construction is a submodule of M [ z , z − 1 ] , although negative filtration indices may occur. Goodness implies that R F ( M ) is a coherent graded R Y -module. Since it is embedded in M [ z , z − 1 ] , multiplication by z is injective. More generally, for the module Rees objects considered here, the only relevant C [ z ] -torsion is absent.
Let R Y , 1 = R Y / ( z − 1 ) and R Y , 0 = R Y / ( z ) ; the first is interpreted after forgetting the grading. There are natural identifications
R Y , 1 ≃ D Y , R Y , 0 ≃ gr ord D Y ≃ Sym O Y T Y .
For π Y : T * Y → Y , the final algebra identifies with π Y * O T * Y ; it is not literally a sheaf on T * Y until the relative spectrum construction is made. Because X is singular, this identification does not supply an unqualified identification with O T * X .
Proposition 21
(Fibers of a good Rees module). For ( M , F ) ∈ Fil Y str , put M ˜ = R F ( M ) . Then
M ˜ ⊗ C [ z ] C [ z ] / ( z − 1 ) ≃ M , M ˜ ⊗ C [ z ] L C [ z ] / ( z ) ≃ gr F M .
The second isomorphism is in the derived category of graded gr ord D Y -modules; the right-hand side is concentrated in degree zero. In particular, Tor 1 C [ z ] ( M ˜ , C [ z ] / ( z ) ) = 0 .
Proof. 
The specialization z = 1 identifies the homogeneous section m z p with m. Exhaustiveness makes this map surjective, and the standard Rees relations identify its kernel with ( z − 1 ) M ˜ . At z = 0 , the degree-p component of the quotient is
F p M z p / z F p − 1 M z p − 1 ≃ F p M / F p − 1 M .
Finally, the resolution 0 → C [ z ] → z C [ z ] → C [ z ] / ( z ) → 0 computes the derived fiber. Its degree- ( − 1 ) cohomology is the kernel of multiplication by z on M ˜ , which vanishes by the Rees embedding. The same argument applies locally on Y and respects the R Y -module structure. □

4.4. The Attachment Triangle in a Rees Category

Write
A ˜ = R F ( A ) , P ˜ = R F ( P ) , C ˜ = R F ( C ) .
Let D gr b ( R Y ) denote the bounded derived category of graded left R Y -modules in which the coherent Rees objects above and their cones are represented. A strict filtered conflation is sent to a short exact sequence of graded Rees modules; this gives an exact functor on the derived category of the strict exact structure for the objects and maps used here.
Proposition 22
(The Rees connecting morphism). There is a distinguished triangle
A ˜ ⟶ P ˜ ⟶ C ˜ → ϵ R A ˜ [ 1 ]
in D gr b ( R Y ) , induced by (94). Its fiber at z = 1 is the underlying D Y -module extension. Its derived fiber at z = 0 is the triangle associated with the strictly exact sequence of graded modules
0 ⟶ gr F A ⟶ gr F P ⟶ gr F C ⟶ 0 .
In particular, the specialized attaching arrow is
ϵ 0 : = ϵ R ⊗ C [ z ] L C [ z ] / ( z ) .
Proof. 
By strictness, taking F p of (94) gives a short exact sequence for every p. Taking their direct sum, with z p recording the degree, gives
0 ⟶ A ˜ ⟶ P ˜ ⟶ C ˜ ⟶ 0 .
A short exact sequence of graded modules determines the triangle (96). The two specializations are exact on these three Rees modules by Proposition 21. Applying the corresponding derived tensor functors to the triangle gives respectively the underlying differential-module triangle and the associated-graded triangle. Naturality of the connecting morphism identifies their last arrows with the respective specializations of ϵ R . □
Remark 26
(Comparison with the original Yoneda class). The map ϵ Σ is a morphism of mixed Hodge modules in a derived category; ϵ F is its image in the derived category of the strictly filtered exact category; and ϵ R is the image of ϵ F under Rees. These arrows have different source and target categories. Only the specified sequence and its functorial images identify them. Neither the rational costalk class nor an exceptional-curve trace determines ϵ 0 without filtered compatibility.

4.5. A Minimal Rees Test before the Conifold Calculation

We give two explicit tests in which the underlying module extension is identical but the filtered structure differs. These are computations in good filtered modules, not assertions that the examples underlie mixed Hodge modules. Let D = C [ x ] have its polynomial-degree filtration, F p ord D = C [ x ] ≤ p for p ≥ 0 . Put
A = D a / ( x a ) , C = D b ¯ / ( x b ¯ ) , P = ( D a ⊕ D b ) / ( x a , x b − a ) .
Thus A ⊂ P is generated by a, the quotient is generated by the class b ¯ of b, and the underlying sequence
0 ⟶ A ⟶ P ⟶ C ⟶ 0
is nonsplit. Indeed, every lift of b ¯ is b + λ a , while x ( b + λ a ) = a ≠ 0 .
Proposition 23
(Nonzero specialization). Give A, P, and C the quotient filtrations determined by
deg F ( a ) = 1 , deg F ( b ) = 0 , deg F ( b ¯ ) = 0 .
This is a strictly filtered sequence. Put ξ = x z , the degree-one generator of the Rees algebra C [ z , ξ ] . Writing a 1 = a z and b 0 = b , the Rees middle module has relations
ξ a 1 = 0 , ξ b 0 = a 1 .
The specialized extension at z = 0 is nonsplit.
Proof. 
The filtrations on the underlying vector spaces are
F − 1 P = 0 , F 0 P = C b , F p P = P ( p ≥ 1 ) ,
F 0 A = 0 , F p A = A ( p ≥ 1 ) , F − 1 C = 0 , F p C = C ( p ≥ 0 ) .
They are compatible with the degree filtration on D, since x F 0 P ⊆ F 1 P and x F 1 P = C a ⊆ F 2 P . For every p, F p A = A ∩ F p P and the image of F p P in C is F p C ; the sequence is strict. The equations x a = 0 and x b = a yield, on homogeneous Rees generators, ξ a 1 = 0 and ξ b 0 = a 1 . Modulo z, these relations remain unchanged. Any lift of the quotient generator in the specialized middle module is b 0 + λ a 1 with λ ∈ C , and multiplication by ξ sends this lift to a 1 ≠ 0 . No splitting exists. □
Proposition 24
(A nonzero Rees class killed by specialization). Give the same underlying sequence (98) the filtrations determined by
deg F ( a ) = deg F ( b ) = deg F ( b ¯ ) = 0 .
This is again a strictly filtered sequence. Its Rees module has homogeneous relations
ξ a 0 = 0 , ξ b 0 = z a 0 ,
where a 0 = a and b 0 = b . The class of the Rees sequence is nonzero, but its derived specialization at z = 0 is zero. All three Rees modules are z-torsion-free; hence their Tor 1 C [ z ] groups with C [ z ] / ( z ) vanish.
Proof. 
Here F p A = F p P = F p C = 0 for p < 0 and equals the whole respective module for p ≥ 0 . The underlying sequence is exact, and the filtration on P induces precisely the filtrations on A and C, so strictness holds. Since x b = a is an equality in filtration degree 0, multiplying it by z gives the homogeneous Rees relation ξ b 0 = z a 0 .
If the Rees sequence split, specialization at z = 1 would split (98), contradicting its nonsplitting. Thus its Rees connecting class is nonzero. On reduction modulo z, the middle module has relations ξ a 0 = ξ b 0 = 0 , and is therefore the direct sum of the two rank-one quotient modules generated by a 0 and b 0 . Its specialized short exact sequence splits. Finally, each Rees module embeds into the Laurent-polynomial module of its underlying filtered vector space. Therefore multiplication by z is injective on each one. Applying the resolution of C [ z ] / ( z ) used in Proposition 21 proves the asserted Tor vanishing. □
Remark 27
(What the examples prove). The two filtrations give the same nonsplit extension of D-modules but distinct images under associated graded. Consequently, neither strictness nor the underlying nonzero connecting morphism determines specialization survival. The examples are finite tests of the filtered construction, not a substitute for the filtered presentation of the corrected conifold mixed Hodge module.

4.6. Survival, Kernels, and a Calculable Defect

Let D gr ( R Y ) denote a derived category of graded R Y -modules containing the Rees objects above. Let D gr ( R Y , 0 ) denote the corresponding graded derived category for R Y , 0 = gr ord D Y . Derived specialization is the exact functor
sp 0 ( − ) = − ⊗ C [ z ] L C [ z ] / ( z ) .
It induces a homomorphism on morphism groups
Hom D gr ( R Y ) ( C ˜ , A ˜ [ 1 ] ) ⟶ Hom D gr ( R Y , 0 ) ( sp 0 C ˜ , ( sp 0 A ˜ ) [ 1 ] ) , γ ⟼ sp 0 ( γ ) .
Here Hom means degree-zero morphisms in the specified graded derived categories. For a graded object N, we write N ( − 1 ) for the shift characterized by N ( − 1 ) p = N p − 1 , so multiplication by z defines a degree-zero morphism z : N ( − 1 ) → N .
Proposition 25
(Specialization kernel criterion). For A ˜ , C ˜ as above, the kernel of (99) is
ker ( sp 0 ) = im ( Hom D gr ( R Y ) ( C ˜ , A ˜ ( − 1 ) [ 1 ] ) → z ∘ − Hom D gr ( R Y ) ( C ˜ , A ˜ [ 1 ] ) ) .
In particular,
ϵ 0 ≠ 0 ⟺ [ ϵ R ] ∉ im ( z ∘ − ) .
This is a criterion in the Rees-derived category; it makes no assertion that the conifold class satisfies it.
Proof. 
Since z is central in R Y , multiplication by z yields a distinguished triangle in the graded derived category
A ˜ ( − 1 ) → z A ˜ ⟶ sp 0 A ˜ ⟶ A ˜ ( − 1 ) [ 1 ] .
Apply the cohomological functor Hom D gr ( R Y ) ( C ˜ , − ) to (101) and take the portion in cohomological degree one. Exactness gives
Hom ( C ˜ , A ˜ ( − 1 ) [ 1 ] ) → z ∘ − Hom ( C ˜ , A ˜ [ 1 ] ) ⟶ Hom ( C ˜ , ( sp 0 A ˜ ) [ 1 ] ) .
By derived extension-of-scalars adjunction, the last group is naturally
Hom D gr ( R Y , 0 ) ( sp 0 C ˜ , ( sp 0 A ˜ ) [ 1 ] ) .
The middle-to-right arrow is the map on Hom groups induced by specialization. The kernel-image identity follows from exactness of the displayed sequence. Applying the identity to [ ϵ R ] proves the final assertion. □
Corollary 3
(A sufficient detection condition). If
Hom D gr ( R Y ) ( C ˜ , A ˜ ( − 1 ) [ 1 ] ) = 0 ,
then the map (99) is injective. More generally, it is injective on the subgroup generated by [ ϵ R ] if that class has no preimage under multiplication by z.
Proof. 
Both assertions follow directly from (100). □
Remark 28
(No automatic passage from MHM to coherent extensions). The class [ ϵ R ] is defined by the strict filtered sequence, whereas [ ϵ 0 ] lies in a derived category of coherent symbol modules. A nonzero class in Ext MHM ( X ) 1 ( R Σ H , I H ) may become zero after forgetting rational and weight structures or after specializing at z = 0 . Proposition 25 identifies the latter source of vanishing for theReesclass without asserting that the former forgetful map is injective.

4.7. The General Decision Criterion and Its Conifold Resolution

Definition 21
(Local specialization datum). For p ∈ Σ , let Y p ⊂ Y be a sufficiently small algebraic neighborhood on which the filtered components of the corrected extension admit presentations. A local specialization datum consists of the restricted Rees objects A ˜ p , P ˜ p , C ˜ p , the maps in their strict exact sequence, and either a finite graded R Y p -resolution of C ˜ p or another explicit model computing
RHom R Y p ( C ˜ p , A ˜ p ) .
The datum includes filtration degrees, differential maps, and the chosen embedding X ↪ Y .
Proposition 26
(Finite local decision procedure). Assume a local specialization datum at p includes a bounded complex L • of finite graded projective R Y p -modules representing C ˜ p . Let γ p be the degree-one cohomology class of the morphism complex
Hom R Y p ( L • , A ˜ p )
representing the restricted Rees extension. Then the specialized class vanishes if and only if its reduction in
H 1 Hom R Y p ( L • , A ˜ p ) ⊗ C [ z ] L C [ z ] / ( z )
is zero, with the natural base-change identification whenever the indicated finite projective model provides it. Equivalently, one may test whether γ p lies in the image of the multiplication-by-z morphism in (100) after restriction to Y p .
Proof. 
The projective resolution computes the Rees-derived Hom. Derived specialization of the representing arrow is obtained by applying − ⊗ C [ z ] L C [ z ] / ( z ) to its chain representative. The resulting arrow is null in the specialized derived category exactly when its representing cocycle is a boundary in the specialized morphism complex. Under the stated finite projectivity and base-change identification, the displayed complex computes that morphism group. Independently of this model, the same condition is characterized by Proposition 25. □
Remark 29
(Scope of the conifold assertion). Propositions 20 and 22 prove existence and exactness of the filtered and Rees extensions, and Proposition 25 provides the general necessary-and-sufficient criterion for survival. The local datum needed to decide the determinantal case is now supplied by the localization identification and explicit Hodge-filtration calculation in Theorems 2 and 3 below. The latter proves non-splitting at z = 0 for every nonzero corrected local ODP class; this stronger conclusion does not follow from strictness alone. No Hecke–Rees interchange theorem is assumed here.

4.8. Localization Model and a Non-Split Conifold Specialization

We now evaluate the criterion of Proposition 25 on the actual canonical conifold extension. Let Y = A C 4 with coordinates ( x , y , v , w ) , g = x w − y v , X = { g = 0 } , and o = ( 0 , 0 , 0 , 0 ) . Write ι : X ↪ Y , j : X ∖ { o } ↪ X and u : Y ∖ X ↪ Y . Every Hodge module in this subsection has rational coefficients, although its filtered left- D Y component is a complex-analytic/algebraic D Y -module.
Theorem 2
(The canonical middle Hodge module as a localization quotient). Let
P ˜ loc H : = coker MHM ( Y ) Q Y H [ 4 ] ⟶ H 0 p u * Q Y ∖ X H [ 4 ] ( 1 ) .
Here the arrow is the open-localization unit and the Tate twist is fixed by the residue normalization. Under Kashiwara equivalence for objects supported on X, this is the canonical middle Hodge module
P ˜ loc H ≃ ι * P can H , P can H = H 0 p ( j * j * I H ) .
Its underlying filtered left D Y -module is the Hodge localization quotient
O Y [ g − 1 ] / O Y , F ( 1 ) .
The identification is canonical for the open-localization data; it does not assert equality of this middle term with the independentlymarkedcorrected sequence.
Proof. 
Since u is the complement of a Cartier divisor in smooth Y, open localization of the perverse constant Hodge module is perverse in the indicated affine-open normalization. The inclusion of the constant module into its localization is strict, and the localization quotient is supported on X. The left differential module of this quotient is O Y [ g − 1 ] / O Y ; the residue Tate twist is compensated by the displayed ( 1 ) . On U = X ∖ { o } the smooth divisor residue sequence identifies its restriction with the threefold perverse constant module Q U H [ 3 ] . Moreover i o ! u * = 0 by open–closed recollement on Y, whereas purity of the smooth fourfold gives
i o ! Q Y H [ 4 ] ≃ Q o H ( − 4 ) [ − 4 ] .
Applying i o ! to the localization triangle and then the twist ( 1 ) yields
i o ! P ˜ loc H ≃ Q o H ( − 3 ) [ − 3 ] .
(The cohomological degree is + 3 .) Upon restriction to the closed hypersurface, the recollement triangle is
i * i ! P loc H ⟶ P loc H ⟶ j * j * P loc H ⟶ i * i ! P loc H [ 1 ] .
Its degree-zero perverse cohomology identifies P loc H with H 0 p ( j * Q U H [ 3 ] ) . This is precisely P can H of Lemma 2, with the canonical identity on U. No scalar for the corrected variation-cone class has been chosen or inferred. □
Proposition 27
(Verdier-dual description of the canonical extension). For the determinantal ordinary double point in the conventions of Theorem 2, there is an identification of Hodge modules
P can H ≃ D X ( Q X H [ 3 ] ) ( − 3 ) .
The Verdier dual of the canonical forward extension, after untwisting by ( − 3 ) , is the short exact sequence
0 ⟶ i o * Q o H ( − 1 ) ⟶ Q X H [ 3 ] ⟶ I H ⟶ 0 .
The punctual ( − 1 ) in this dual sequence is not the ( − 2 ) punctual quotient of the forward extension. The statement identifies the canonical, not the independently marked corrected, middle term.
Proof. 
The local hypersurface X is a three-dimensional local complete intersection. Its perverse-normalized constant complex Q X H [ 3 ] lies in the MHM heart for the ordinary-double-point model. The standard open–closed perverse truncation gives Q X H [ 3 ] ≃ H 0 p ( j ! Q U H [ 3 ] ) . Verdier duality exchanges j ! with j * and preserves the perverse heart. Since the intersection Hodge module is pure of weight three and satisfies D X I H ≃ I H ( 3 ) , the residue normalization of Theorem 2 identifies the dual of the perverse constant object, after the indicated ( − 3 ) twist, with H 0 p ( j * Q U H [ 3 ] ) = P can H . Dualizing 0 → I H → P can H → i o * Q o H ( − 2 ) → 0 reverses the exact sequence. The dual of its punctual endpoint is i o * Q o H ( 2 ) ; tensoring the entire dual sequence with Q H ( − 3 ) converts this to i o * Q o H ( − 1 ) , and the middle term becomes Q X H [ 3 ] . Self-duality of I H with its weight-three twist gives the last endpoint and proves the second displayed sequence. The construction is functorial for the specified duality identifications; it does not calibrate the independently marked corrected extension. □
Lemma 6
(First relevant Hodge levels). For the ordinary double point g = x w − y v the reduced Bernstein–Sato polynomial is s + 2 and the minimal exponent is α ˜ g = 2 . With the twist in (104) and the left-module filtration conventions of §Section 4.1, the initial Hodge-filtration levels on the localization quotient at o are
F 1 P can = O Y , o g − 1 + O Y , o O Y , o , F 2 P can = O Y , o g − 2 + O Y , o g − 1 O Y , o .
Consequently
gr 2 F P can ≅ O Y , o / ( g ) .
The punctual quotient has a degree-two Hodge generator: gr 2 F R o H ≅ C o , annihilated by the maximal ideal ( x , y , v , w ) .
Proof. 
For a nondegenerate quadratic form in four variables, the Bernstein–Sato polynomial is ( s + 1 ) ( s + 2 ) ; hence its reduced polynomial is s + 2 and its minimal exponent is 2. For a reduced divisor D = ( g = 0 ) the Hodge ideals satisfy F k O Y ( * D ) = I k ( D ) O Y ( ( k + 1 ) D ) , and the minimal-exponent criterion gives I k ( D ) o = O Y , o if and only if k + 1 ≤ α ˜ g . See [20], together with the reduced-divisor formulation of the Mustaţă–Popa–Saito criterion. Here k = 0 , 1 satisfy the inequality. Thus the Hodge and pole-order filtrations coincide at those two levels; passing to the localization quotient and applying the specified twist ( 1 ) gives (107). Modulo the simple-pole level, the double-pole generator has annihilator exactly ( g ) , giving (108). For the point Hodge module Q o H ( − 2 ) , left filtered direct image across codimension four places the delta generator in degree two, as calculated independently in Lemma 7 below. Its degree-two symbol is therefore C o with trivial coordinate action. □
Theorem 3
(Non-splitting of the specified conifold Rees specialization). The strictly exact canonical sequence 0 → I H → P can H → R o H → 0 has anon-splitassociated-graded sequence in the category of graded gr ord D Y -modules. Equivalently its specialized Rees connecting arrow
ϵ 0 , can : gr F R o ⟶ ( gr F I ) [ 1 ]
is nonzero. The same conclusion holds for everynonzerocorrected ordinary-double-point class with the given quotient Q o H ( − 2 ) and fixed endpoint identifications, independently of the marked scalar λ o ∈ Q × .
Proof. 
Saito strictness yields a short exact sequence of associated-graded modules. If it split, its section s : gr F R o → gr F P can would send the degree-two delta generator δ ¯ to an element
t ∈ gr 2 F P can ≅ O Y , o / ( g )
with nonzero image δ ¯ under the quotient map. Since every coordinate kills δ ¯ , graded gr D Y -linearity implies
x t = y t = v t = w t = 0 in O Y , o / ( g ) .
But g = x w − y v is irreducible, so the local ring O Y , o / ( g ) is a domain, and x is nonzero in it. Thus x t = 0 forces t = 0 , contradicting the required nonzero image. The graded extension is nonsplit, and its derived connecting arrow is nonzero.
At the node the rational Hodge-module extension group is the line computed in Corollary 2. A nonzero corrected class differs from the canonical class by an invertible rational scalar applied to the punctual endpoint. The corresponding change of endpoint marking induces a filtered Hodge-module automorphism and hence an automorphism of the associated-graded endpoint. Splitting is preserved under such an invertible pullback/pushout. Therefore the corrected specialized extension is also nonsplit. This argument does not identify the independently marked arrows. □
Remark 30
(Relationship with the Koszul cocycle). The four-component cocycle of Proposition 29 below represents the same strict Rees extension after choosing a degree-two lift of its punctual generator. The failure of the associated-graded splitting shown above implies that this cocycle cannot become a Koszul coboundary after reduction modulo z. Thus the localization model resolves the decision problem without replacing the four-map construction by an abstract dimension count. The conclusion is specific to the chosen canonical localization model and its nonzero corrected scalar multiples; it is not a general consequence of strictness for arbitrary filtered modules.

4.9. An Explicit Local Rees Complex for the Punctual Endpoint

The preceding criterion can be reduced, for the determinantal ordinary double point, to a finite complex with explicitly given differentials. This reduction uses the actual punctual Hodge module, not a model extension selected after taking the associated graded. Let Y = A C 4 , with coordinates x = ( x 1 , x 2 , x 3 , x 4 ) = ( x , y , v , w ) , and let o = ( 0 , 0 , 0 , 0 ) . Write D = Γ ( Y , D Y ) for the fourth Weyl algebra, and put
R D = R F ord ( D ) , ξ i = z ∂ x i , [ ξ i , x j ] = δ i j z ,
where x i has Rees degree zero and z , ξ i have degree one. For a graded module, 〈 m 〉 denotes a grading shift which places a distinguished free generator in degree m.
Lemma 7
(Punctual filtered presentation and its shifts). Let δ o be the left D-module underlying the direct image of Q o H ( − 2 ) from the point. With the left-module Hodge filtration and the ambient dimension four,
δ o ≃ D / ∑ i = 1 4 D x i , F p δ o = ⨁ | ν | ≤ p − 2 C ∂ ν δ ,
where the sum is zero for p < 2 and ∂ ν = ∏ i ∂ x i ν i . Its Rees module and its derived z = 0 fiber have presentations
R ˜ o ≃ R D / ∑ i R D x i 〈 2 〉 , gr F δ o ≃ C [ ξ 1 , … , ξ 4 ] 〈 2 〉 .
The support of the latter graded module in T * Y is the full cotangent fiber T o * Y .
Proof. 
As an unfiltered left module the direct image from the point is the delta module generated by δ with x i δ = 0 . Poincaré–Birkhoff–Witt normal ordering gives the C -basis ∂ ν δ , hence the first quotient presentation. The left filtered direct-image formula for a smooth closed embedding of codimension four is, in adapted coordinates,
F p ( i o + M ) = ∑ ν ∈ N 4 F p − | ν | − 4 M ∂ ν .
This is the left-module version of Saito’s filtered direct-image construction; the codimension shift is essential [6,8,9]. For the point Hodge structure Q H ( − 2 ) , F q Q H ( − 2 ) = 0 if q < − 2 , and it equals Q if q ≥ − 2 . Substitution into (113) gives exactly the filtration in (111). Taking its Rees module gives the quotient of R D by the homogeneous left relations x i δ = 0 , with δ in degree two. The Rees PBW basis consists of the monomials ξ ν δ over C [ z ] , so this quotient is z-torsion-free and agrees with the Rees module of the specified Hodge filtration. Modulo z, the commutators in (110) vanish, and the quotient by the coordinates x i is C [ ξ 1 , … , ξ 4 ] with the stated shift. Its support is x 1 = … = x 4 = 0 with unrestricted cotangent coordinates. □
Proposition 28
(A finite free Rees resolution). Let E = C 4 with ordered basis e 1 , … , e 4 , and let L • be the cochain complex concentrated in degrees − 4 , … , 0 , with
L − q = R D 〈 2 〉 ⊗ C ⋀ q E , 0 ≤ q ≤ 4 ,
and differential, for a basis wedge, defined by
d − q ( P ⊗ e i 1 ∧ … ∧ e i q ) = ∑ a = 1 q ( − 1 ) a − 1 P x i a ⊗ e i 1 ∧ … e ^ i a … ∧ e i q .
Here right multiplication by x i is a morphism of freeleft R D -modules. The augmentation L 0 ↠ R ˜ o is a quasi-isomorphism. The same complex after z = 0 is a free Koszul resolution of gr F δ o over gr ord D .
Proof. 
The four right multiplication operators by x i commute, so the alternating sum in (115) squares to zero: in every term of d 2 , the two orders of removing distinct basis vectors appear with opposite signs and identical product x i x j = x j x i . The PBW normal form for the Rees Weyl algebra gives a locally free right C [ x 1 , … , x 4 ] -module with basis z a ξ ν . Therefore tensoring the ordinary Koszul resolution of C [ x 1 , … , x 4 ] / ( x 1 , … , x 4 ) with R D over this polynomial algebra preserves exactness. The resulting complex is exactly (114)– (115); its degree-zero cokernel is (112). Because every free term and the quotient have zero z-torsion, its derived specialization is represented by reduction of the complex modulo z. There the coordinate functions form a regular sequence in the commutative polynomial ring C [ x 1 , … , x 4 , ξ 1 , … , ξ 4 ] , giving the asserted Koszul resolution. □
Proposition 29
(The local connecting class as a four-component cocycle). Let
0 ⟶ A ˜ o → a R P ˜ o → b R R ˜ o ⟶ 0
be the restriction to the determinantal chart of theactualstrict Rees extension, with A ˜ o the Rees module of the intersection Hodge module. Choose a homogeneous lift m ∈ ( P ˜ o ) 2 of the canonical punctual generator δ ∈ ( R ˜ o ) 2 and set
a i : = a R − 1 ( x i m ) ∈ ( A ˜ o ) 2 , i = 1 , … , 4 .
Then a = ( a 1 , … , a 4 ) satisfies
x i a j − x j a i = 0 ( 1 ≤ i < j ≤ 4 ) .
In the graded derived category, ϵ R , o is represented by a in the degree-one cohomology of
C q = Hom R D ( L − q , A ˜ o ) , d C 0 ( b ) = ( x 1 b , … , x 4 b ) ,
with d C 1 ( a ) i j = x i a j − x j a i . Changing the lift m changes a by a coboundary. The image of the extension at z = 0 is represented by a ¯ = ( a i mod z ) in the corresponding Koszul complex on gr F A o . Consequently
ϵ 0 , o = 0 ⟺ ∃ b ¯ ∈ ( gr F A o ) 2 : a i ¯ = x i b ¯ for all i .
Proof. 
Since b R ( m ) = δ and x i δ = 0 , b R ( x i m ) = 0 . Exactness of (116) gives a unique a i ∈ A ˜ o with a R ( a i ) = x i m . The action of x i has degree zero, so each a i has the homogeneous degree of m, namely two. For i , j ,
a R ( x i a j − x j a i ) = x i x j m − x j x i m = 0 .
Injectivity of a R proves (118). Applying Hom R D ( − , A ˜ o ) to the resolution of Proposition 28 identifies C 0 with the degree-two part of A ˜ o , C 1 with four copies of the same degree, and the first two cochain differentials with those displayed above. The extension (116) is represented by lifting the augmentation generator and measuring the failure of its four relations to hold in the lift, exactly the tuple a . If m is replaced by m + a R ( b ) , then a i is replaced by a i + x i b , the coboundary of b. This verifies that the cohomology class is intrinsic.
The free resolution remains a resolution after derived reduction modulo z, by Proposition 28. Functoriality of the extension class under base change sends the cochain representative a to a ¯ . The latter is a boundary if and only if it lies in the image of the first differential d C 0 in the specialized complex; this is precisely the simultaneous system in (120). □
Remark 31
(The resolved nonvanishing test and its chain representative). The four equations (117)– (120) give a chain-level test for the selected Rees class. The filtered localization presentation of Theorem 2, its first relevant Hodge levels in Lemma 6, and the annihilator argument of Theorem 3 now resolve the zero/nonzero question for the actual canonical conifold class: its reduction is nonzero. The same is true for a nonzero corrected class by invertible endpoint rescaling. The explicit four-component cocycle remains a useful independent chain model, but no longer constitutes an unresolved prerequisite for the survival theorem. The exact marked variation-cone scalar is still not determined.

4.10. A Categorical Boundary Toward KKPY F-Bundles

Theorem 3 establishes a non-split specialization in the graded symbol-module category, not a realization in a category of quantum F-bundles. The filtered localization module is supported on the proper hypersurface X det ⊂ Y and cannot be a locally free flat bundle with nonsingular connection on all of the connected smooth ambient fourfold. This does not rule out a category of integrable Rees modules with singular support; it rules out only the indicated smooth ambient flat-bundle input type. The parameters z and u encode independently defined structures; renaming one does not construct an extension-preserving comparison functor. Such a comparison would require a specified geometric intermediate operation and its action on the marked Yoneda arrow, as well as the relevant connection and Euler structures.
The functor established above has type
D b MHM ( X 0 ) ⟶ D b ( FilRH str ( Y ) ) ⟶ D gr b ( R Y ) ,
for the fixed embedding X 0 ↪ Y . Its second arrow is the exact Rees operation on strictly exact sequences. Its associated-graded specialization has target the derived category of Sym O Y T Y -modules. None of these target categories is, by definition, the KKPY category Fbun k ( B F ) of Definition 7.
A separate filtered-flat Rees construction requires a bundle with an integrable connection on a smooth base and a Griffiths-transverse filtration by subbundles. The objects For F , i ( I H ) and For F , i ( P H ) are generally holonomic modules supported on the singular central fiber and do not have that type. Moreover, the present parameter z is the Rees parameter for the Hodge filtration on the ambient differential module; it is not the auxiliary u parameter of a quantum F-bundle. Therefore the functor (121) cannot be composed with the formal filtered-flat-to-TE construction without an independently defined intermediate operation. The separate formal filtered-flat construction is treated after Section 6; no composition with the singular Rees functor is presumed here.

4.11. One-node Coherent Geometry and the Limits of Spectral Identification

Let Y = A C 4 have coordinates ( x , y , v , w ) , and let
X det = { x w − y v = 0 } ⊂ Y .
Its singular locus is the origin o, and its smooth locus is U det = X det ∖ { o } . Denote by ι : X det ↪ Y the embedding and by M I the left filtered differential module underlying ι * IC X det H . The associated graded gr F M I is a coherent module over Sym O Y T Y , hence corresponds to a coherent sheaf on T * Y .
Definition 22
(Conormal and characteristic supports). Let g = x w − y v . The conormal variety to the smooth locus U det , regarded as a subset of Y, is the closure
Λ det = ( a , λ d g a ) : a ∈ U det , λ ∈ C ¯ ⊂ T * Y .
For a good filtered holonomic D Y -module M, its characteristic variety is
Char ( M ) = Supp T * Y ( gr F M ) ,
which is independent of the choice of good filtration as a closed subset.
Proposition 30
(Explicit conormal equations). Let ( ξ x , ξ y , ξ v , ξ w ) be the cotangent fiber coordinates dual to ( x , y , v , w ) . On U det the conormal is given parametrically by
x w − y v = 0 , ( ξ x , ξ y , ξ v , ξ w ) = λ ( w , − v , − y , x ) , λ ∈ C .
Its closure is Λ det . In particular, the conormal construction alone does not determine the multiplicity or the full coherent module gr F M I .
Proof. 
Differentiation gives
d g = w d x − v d y − y d v + x d w .
At each point of U det , d g is nonzero, so the conormal line is generated by this covector. Taking all points and the Zariski closure gives the asserted parameter presentation. A support computation records where a module can be nonzero, but it does not determine its module presentation, characteristic-cycle multiplicities, or an extension class. These require the good filtered presentation. □
Remark 32
(Three distinct spectral constructions). The characteristic support in T * Y is associated with an algebraic filtered differential module. A Hitchin spectral curve is defined by the characteristic polynomial of a Higgs field over a smooth base curve; the critical-value cover of the stabilized A 2 family in (24) is T 2 = 4 9 s 3 . These objects have different ambient spaces and defining data. Neither the presence of an exceptional P 1 in a small resolution of X det nor agreement of selected spectral dimensions constructs a comparison functor among them. The BNR and nonabelian-Hodge correspondences have their own smoothness, stability, and moduli hypotheses; they are not invoked here as an extension-preserving identification [21,22,23].

5. Framed Minuscule Hecke Functors and Attachment-Level Intertwining

This section establishes the geometric and coefficient-theoretic input for one lower minuscule modification of the corrected conifold attaching morphism. The operation is not a numerical correspondence between two projective lines: it consists of an evaluation map, a scheme of invariant lattices, a genuine modified-Higgs target map on its admissible domain, a coefficient kernel, and a pull–tensor–push functor on specified derived categories. We construct the invariant-line equation and its base changes explicitly. In particular, we separate the full invariant-line incidence from the kernel-line small resolution of the determinantal conifold; conflating these spaces would incorrectly transfer their fibers and their Hodge-module direct images.
The corrected conifold extension and the localized one-step survival results are imported from [2,4]. We give chart-level constructions of the corresponding maps, distinguish the two incidence components, calculate the first-jet transformation and an auxiliary exceptional-fiber direct image, and record the induced morphisms and perverse long exact sequences in the relevant derived categories. The exceptional-curve Gysin identification proved in Section 3.9 and Appendix A.8 concerns the kernel-line small resolution; it does not identify that resolution with the full invariant-line or genuine modified-Higgs correspondence. These constructions furnish the one-step input to Section 6, without invoking an unconstructed spectral or atomic F-bundle functor.
We use the conventions of Section 2, particularly the exact sequence (14), its connecting arrow (15), and the distinction between the punctual ( − 2 ) -twisted quotient and the nearby-cycle interface. All pullbacks and pushforwards below are derived operations unless explicitly restricted to an abelian heart. Their existence is supplied, for finite-type algebraic schemes, by Saito’s six-operation formalism and its enhanced form [6,10]. No assertion about arbitrary quotient stacks follows solely from this scheme-level statement.

5.1. The Lattice Model for G L 2

Let C be a smooth complex projective curve, x ∈ C a closed point, and w a uniformizing parameter at x. Set
O = C [ [ w ] ] , K = C ( ( w ) ) , Λ 0 = O 2 , V = Λ 0 / w Λ 0 .
Thus V is a two-dimensional complex vector space.
Definition 23
(Lower minuscule modification). A lower minuscule lattice is a submodule Λ ⊂ Λ 0 over O such that
w Λ 0 ⊂ Λ ⊂ Λ 0 , dim C ( Λ 0 / Λ ) = 1 .
The moduli functor of such modifications is understood on families for which the corresponding quotient is locally free of rank one.
Proposition 31
(The minuscule parameter space). The scheme of lattices satisfying (122) is naturally isomorphic to P ( V ) ≃ P 1 , where P ( V ) parametrizes one-dimensional vector subspaces of V.
Proof. 
Let ρ : Λ 0 → V be reduction modulo w. Since w Λ 0 ⊂ Λ , the image ℓ = ρ ( Λ ) is a C -subspace of V, and
Λ 0 / Λ ≃ V / ℓ .
The dimension condition therefore gives dim C ℓ = 1 . Conversely, for every line ℓ ⊂ V , the inverse image ρ − 1 ( ℓ ) contains w Λ 0 and has colength one. The constructions are inverse to one another and commute with base change for locally free rank-one quotients; hence they define the claimed isomorphism of schemes. □
Remark 33
(Universal line and the family-level construction). Let S be a complex scheme and let L ⊂ ( Λ 0 / w Λ 0 ) ⊗ O S be an invertible rank-one direct summand whose quotient is locally free. Its inverse image Λ S in Λ 0 ⊗ ^ O S satisfies an exact sequence of O S -modules
0 ⟶ Λ S ⟶ Λ 0 ⊗ ^ O S ⟶ ( ( Λ 0 / w Λ 0 ) ⊗ O S ) / L ⟶ 0 .
Conversely such a locally free colength-one quotient determines L . The two assignments respect pullback in S, which is the scheme-theoretic part of the claimed projective-space identification. The family is a moduli functor of lower lattices, not yet a functor on Hodge-module coefficients.
Let φ be a Higgs field. In a local trivialization of the bundle, its action on the fiber at x is represented by a matrix A ∈ Mat 2 ( C ) . Compatibility of the elementary modification with a holomorphic Higgs field imposes the condition A ℓ ⊂ ℓ for the corresponding line, with any further pole or residue conditions retained from the chosen Higgs-modification problem.
Lemma 8
(Invariant-line equation). Write
A = a b c d , ℓ = C ( s , t ) ∈ P 1 .
Then A ℓ ⊂ ℓ if and only if
c s 2 + ( d − a ) s t − b t 2 = 0 .
On the chart s ≠ 0 , with u = t / s , the equation becomes c + ( d − a ) u − b u 2 = 0 .
Proof. 
A line is invariant precisely when ( s , t ) and A ( s , t ) are linearly dependent. Their determinant equals
det s a s + b t t c s + d t = c s 2 + ( d − a ) s t − b t 2 .
Dividing by s 2 on the indicated chart gives the second assertion. □
Example 1
(A doubled invariant line is not a doubled kernel line). For A = 0 1 0 0 , Equation (123) is − t 2 = 0 . The full invariant-line scheme r − 1 ( A ) is the length-two point supported at [ 1 : 0 ] , whereas the kernel-line fiber ρ ker − 1 ( A ) is the single reduced line ker A . Thus even when their underlying invariant lines agree, the scheme-theoretic fibers—and therefore their coherent coefficient theories—can differ.
Remark 34
(Geometric versus categorical identification). The parameter space P ( V ) and the exceptional curve of a small resolution of the determinantal conifold are both projective lines. This identifies neither their ambient spaces nor the categories of mixed Hodge modules on them. Any comparison of attaching morphisms requires a specified geometric map, coefficient functor, and normalization; see [4,5].

5.2. Determinantal Geometry and the Framed Invariant-Line Correspondence

Put M 2 = Mat 2 ( C ) and let
X det = { A ∈ M 2 : det A = 0 } .
Introduce the incidence variety
I = { ( A , ℓ ) ∈ M 2 × P 1 : A ℓ ⊂ ℓ } ,
with projections r : I → M 2 and λ : I → P 1 . Set I 0 = r − 1 ( X det ) , and write r 0 : I 0 → X det for the restriction. The map r is projective, being the restriction of the projection M 2 × P 1 → M 2 to a closed subscheme.
Lemma 9
(Incidence chart and determinant factorization). On the chart ℓ = C ( 1 , u ) , the coordinate ring of I is defined by
c + ( d − a ) u − b u 2 = 0 .
The eigenvalue of A on ℓ is ξ = a + b u , and the induced eigenvalue on C 2 / ℓ is ζ = d − b u . Consequently,
det A = ξ ζ , I 0 ∩ { s ≠ 0 } = { ξ ζ = 0 } inside I ∩ { s ≠ 0 } .
Proof. 
The invariant-line equation implies A ( 1 , u ) = ( a + b u ) ( 1 , u ) , giving ξ . In the basis ( ( 1 , u ) , ( 0 , 1 ) ) , the matrix of A is upper triangular, with diagonal entries ξ and tr ( A ) − ξ = d − b u . Its determinant is their product. Restriction to X det gives the equation ξ ζ = 0 . □
Proposition 32
(The two incidence components and the small resolution). Let I 0 = r − 1 ( X det ) as above and put
J ker = { ( A , ℓ ) : A ℓ = 0 } , J im = { ( A , ℓ ) : im A ⊆ ℓ } .
Both are closed in X det × P 1 . Scheme-theoretically, I 0 = J ker ∪ J im , with the scheme-theoretic union defined locally by ξ ζ = 0 . The projection
ρ ker : J ker ⟶ X det
is a projective small resolution, with exceptional fiber P 1 over A = 0 . In contrast, r 0 : I 0 → X det is generically two-to-one on the locus of rank-one matrices with nonzero trace and is therefore not a small resolution.
Proof. 
In the chart ℓ = C ( 1 , u ) , the invariance equation allows u , b , ξ = a + b u , ζ = d − b u as independent coordinates, since
a = ξ − b u , d = ζ + b u , c = ( ξ − ζ ) u − b u 2 .
Consequently the coordinate ring of I 0 in this chart is C [ u , b , ξ , ζ ] / ( ξ ζ ) . The equation ξ = 0 means A ( 1 , u ) = 0 ; it defines J ker , whereas ζ = 0 means the induced action of A on C 2 / ℓ vanishes and hence im A ⊆ ℓ . The intersection of the two ideals is ( ξ ζ ) , proving the claimed union, including its scheme structure. The argument is invariant under change of projective-line chart.
For each line ℓ, a matrix killing ℓ is exactly a map C 2 / ℓ → C 2 . Writing L = O P 1 ( − 1 ) and Q = C 2 ⊗ O P 1 / L ≃ O P 1 ( 1 ) , we have
J ker ≃ Tot ( Hom ( Q , O ⊕ 2 ) ) = Tot ( O ( − 1 ) ⊕ 2 ) .
It is smooth of dimension three. Since it is closed in X det × P 1 , its projection is projective. Every nonzero rank-one matrix has a unique kernel line; over zero every line is a kernel. Thus ρ ker is birational and its exceptional locus is a curve in a smooth threefold, of codimension two, proving smallness. The codimension-two regular embedding of the exceptional curve also supports the purity/Gysin construction in Appendix A.8; that construction uses ρ ker , not r 0 . Finally, a rank-one matrix with nonzero trace has distinct zero and nonzero eigenlines. Both are invariant, giving two points of r 0 − 1 ( A ) , whereas ρ ker − 1 ( A ) contains only its zero eigenspace. This proves the distinction. □
Corollary 4
(An explicit exceptional-fiber coefficient image). Let o = 0 ∈ X det , let i o : { o } ↪ X det , and let Q o = i o * Q H ( − 2 ) in MHM ( X det ) . For E = ρ ker − 1 ( o ) ≃ P 1 , proper base change gives an isomorphism in D b MHM ( { o } )
i o * ρ ker * ρ ker * Q o ≃ R Γ ( E , Q E H ) ( − 2 ) .
Its nonzero cohomology Hodge structures are Q H ( − 2 ) in degree zero and Q H ( − 3 ) in degree two. This is astalkdirect-image calculation for the pulled-back punctual coefficient. It is distinct from the extraordinary-pullback Gysin arrow of Appendix A.8: proper base change computes the displayed endpoint objects, but does not thereby identify a connecting morphism. It is also not an identification of ρ ker with the actual target map q + .
Proof. 
The Cartesian square with i o and the closed exceptional fiber, followed by proper base change, identifies ρ ker * i o * Q H ( − 2 ) with the pullback of the punctual coefficient to E, extended by zero, and its pushforward with R Γ ( E , Q E H ( − 2 ) ) . The rational cohomology of P 1 is concentrated in degrees zero and two. Its Hodge structures are Q H and Q H ( − 1 ) , respectively; the fixed ( − 2 ) twist gives the two stated groups. No t-exactness of ρ ker * is assumed. The scheme-theoretic exceptional fiber and its constant Hodge cohomology are distinct from the local costalk of the conifold intersection complex and from the connecting map between those endpoint objects. □
Let M − fr be a finite-type framed algebraic chart of the Higgs-modification source considered in [4], and let
e : M − fr ⟶ M 2
be its algebraic evaluation map. Define the framed incidence pullback by the Cartesian square
H μ fr → h I p − ↓ p − r ↓ r M − fr → e M 2 .
Let q + : H μ fr → M + fr denote the target map of the selected admissible Higgs modification, wherever that target map has been constructed. The existence and domain of q + are separate from the existence of the Cartesian square (125).
To pull the corrected conifold attachment back along an evaluation map, one must specify the conifold locus. Write
M − , 0 fr = e − 1 ( X det ) , e 0 : M − , 0 fr → X det ,
and set H μ , 0 fr = H μ fr × M 2 X det . The restricted diagram supplies the source incidence for the central-fiber modification. It becomes a correspondence to the modified-Higgs target only on the admissible chart where the latter is defined; denote the restricted maps there by p − and q + and the target by M + , 0 fr . The condition that p − is the base change of a projective morphism does not establish properness of q + , a distinction retained in the coefficient operations below.
Lemma 10
(The genuine first-step chart and its jet restriction). Fix the first lower lattice Λ 1 = w O e 1 ⊕ O e 2 , and let a holomorphic Higgs field in the original frame have matrix
A ( w ) = a ( w ) b ( w ) c ( w ) d ( w ) .
The compatibility condition A ( w ) Λ 1 ⊆ Λ 1 is exactly b ( 0 ) = 0 . Writing b ( w ) = w β ( w ) , the transformed field in the basis ( w e 1 , e 2 ) is
A 1 ( w ) = a ( w ) β ( w ) w c ( w ) d ( w ) , A 1 ( 0 ) = a 0 b 1 0 d 0 ,
where a 0 = a ( 0 ) , d 0 = d ( 0 ) and b 1 is the coefficient of w in b. Thus the transformedresiduedepends on the original first jet. Its full first jet need not be determined by the original first jet alone.
Proof. 
We calculate A ( w e 1 ) = w a e 1 + w c e 2 and A ( e 2 ) = b e 1 + d e 2 . The second column belongs to Λ 1 precisely when b ∈ w O . Relative to the adapted basis these columns give the displayed A 1 ( w ) . Reduction modulo w yields A 1 ( 0 ) , since β ( 0 ) = b 1 . For example the coefficient of w in β = b / w is the coefficient b 2 of w 2 in b, which the original first jet does not determine. Therefore this chart formula does not define a self-map of the entire first-jet parameter space with its first-jet target structure. □
Remark 35
(The two geometric arrows in the framed square). The projection p − in (125) is obtained by base-changing the projective map r; hence p − is projective. The different arrow q + records the modified Higgs pair after the lattice change. On the adapted formal chart its field is given by (126), but its global/algebraic existence and admissible domain are those of the chosen Hecke-modification construction in [4]. Projectivity of p − alone does not establish properness of q + , and the formula for A 1 ( 0 ) does not define a global target map on truncated jets: the coefficient of w in A 1 ( w ) may depend on the original second jet.
Definition 24
(Evaluated source attachment). Let
I det H ⟶ P det H ⟶ R det H → ϵ det I det H [ 1 ]
be the specified determinantal-conifold triangle of [2,4]. Define the derived source objects and morphism by
I − H = e 0 * I det H , P − H = e 0 * P det H , R − H = e 0 * R det H , ϵ − = e 0 * ( ϵ det ) .
When the evaluation is smooth of pure relative dimension d, the perverse-normalized pullback is e 0 * [ d ] ; if extraordinary pullback is used, smooth purity identifies it with e 0 ! ( − d ) [ − d ] in the present Tate convention. These are separately normalized derived operations. Their common shifts and twists must be applied to all three endpoints and the connecting arrow, not only to the punctual quotient. The costalk-detection statement of Proposition 12 may be used only when its actual smoothness hypothesis holds.
Remark 36
(Three distinct direct images). The auxiliary line-parameter projection of [4], denoted q line , is not the target map q + of the genuine Higgs modification. Its operations q line * and q line ! are also distinct whenever that auxiliary map is nonproper. Vanishing under the former does not imply vanishing under the latter or under q + * . The localized *- and !-transport statements on the admissible target opens U and U ′ ⊂ U are used only with the maps and hypotheses in [4].
Remark 37
(Bounded convolution slice). The Springer/convolution-slice comparison established in the restricted geometry of [3,4] is an input. It is not a global isomorphism of the Higgs-modification correspondence with the affine-Grassmannian convolution correspondence. In particular, even the canonical boundary–Gysin equality now proved under the projective small-resolution hypothesis in Theorem 1 concerns a morphism i o * Q H ( − 2 ) → I det H [ 1 ] on the conifold. It gives no identification of the full invariant-line incidence with the small resolution and does not specify a monoidal functor on Higgs or spherical coefficient categories. The corrected marked scalar λ o remains independent of that canonical comparison.

5.3. The Derived Pull–Push Operator

Let X and Y be separated finite-type complex schemes and let X ← p Z μ → q Y be a correspondence of the same type as the restricted framed diagram above. Write D H ( T ) for the bounded stable enhancement of D b MHM ( T ) , for T = X , Z μ , Y . A kernel is an object K μ ∈ D H ( Z μ ) . The tensor symbol below denotes the derived tensor product in the six-functor category, with the cohomological and Tate normalizations included in the choice of K μ .
Definition 25
(Star Hecke operator). Assume that the following expression preserves the chosen bounded coefficient subcategories. Define
H μ ( − ) = q * K μ ⊗ L p * ( − ) : D H ( X ) ⟶ D H ( Y ) .
For the proper-support convention used in Section 6, we write H μ , ! * ( M ) = q ! ( K μ ⊗ L p * M ) . The separately normalized exceptional-pullback operator is H μ , ! ! ( M ) = q ! ( K μ ⊗ L p ! M ) . These are not interchangeable. Properness of q identifies its *- and !-pushforwards on the same coefficients, whereas changing p * to p ! requires an independent purity or duality comparison with its shifts and Tate twists.
Proposition 33
(Exact transport of a marked arrow). Under the hypotheses of Definition 25, H μ is an exact functor of stable categories. The source triangle from Definition 24 is sent to the distinguished triangle
H μ ( I − H ) ⟶ H μ ( P − H ) ⟶ H μ ( R − H ) → H μ ( ϵ − ) H μ ( I − H ) [ 1 ] .
The connecting arrow is uniquely specified by applying the functor to the chosen arrow ϵ − ; no new generator of an extension space is selected.
Proof. 
The functors p * , − ⊗ L K μ , and q * are exact operations in the stable six-functor setting. Their composite preserves cofiber and fiber sequences. Applying it to I − H → P − H → R − H → ϵ − I − H [ 1 ] therefore gives (128). Compatibility of the last arrow with the suspension equivalence is part of exact functoriality. No assertion of perverse t-exactness or nonvanishing is used. □
Proposition 34
(Perverse long exact sequence and the transported boundary). In the setting of Proposition 33, write A + = H μ ( I − H ) , P + = H μ ( P − H ) , C + = H μ ( R − H ) , and ϵ + = H μ ( ϵ − ) . For every integer n, cohomology in the standard t-structure on D b MHM ( Y ) yields
… ⟶ H n − 1 ( C + ) → δ n − 1 H n ( A + ) ⟶ H n ( P + ) ⟶ H n ( C + ) → δ n H n + 1 ( A + ) ⟶ … ,
where δ n = H n ( ϵ + ) , using the canonical shift identification. In particular, there is a natural short exact sequence in MHM ( Y ) ,
0 ⟶ coker ( δ n − 1 ) ⟶ H n ( P + ) ⟶ ker ( δ n ) ⟶ 0 .
A nonzero δ n detects ϵ + ≠ 0 , but a vanishing collection of δ n does not in general prove the derived arrow is zero.
Proof. 
Apply H n to the distinguished triangle (128) and use H n ( A + [ 1 ] ) = H n + 1 ( A + ) . The connecting homomorphism is induced by the given  ϵ + ; it is not an abstract map chosen afterward. Exactness identifies the kernel of H n ( A + ) → H n ( P + ) with im δ n − 1 and the image of H n ( P + ) → H n ( C + ) with ker δ n . Taking the corresponding quotient gives (130). If ϵ + = 0 , all its images under cohomological functors vanish; the converse need not hold because cohomological functors need not be faithful on morphisms of derived objects. □
Remark 38
(Perverse placement and detection). The image objects in (128) may have nonzero perverse cohomology in several degrees. Thus H μ ( ϵ − ) is a derived morphism, not necessarily a Yoneda class in Ext MHM ( Y ) 1 between two perverse-heart objects. Nonvanishing on a designated open subset of the unframed target, and the degree-two attachment described on U ′ , are separate geometric inputs from [4]; they do not follow from Proposition 33.
Lemma 11
(Local coefficient test for the actual target map). Assume q + : H μ , 0 fr → M + , 0 fr is proper on the support of the chosen coefficient complexes. For a geometric point y of the target, write F y = q + − 1 ( y ) and p y = p − | F y . Proper base change identifies
i y * H μ ( R − H ) ≃ R Γ F y , ( K μ | F y ) ⊗ p y * R − H , i y * H μ ( I − H ) ≃ R Γ F y , ( K μ | F y ) ⊗ p y * I − H .
Under these identifications the target-stalk attaching morphism is R Γ ( F y , K μ | F y ⊗ p y * ϵ − ) . Hence a nonzero value of that actual morphism proves H μ ( ϵ − ) ≠ 0 . For the q + ! operator the corresponding stalk base-change formula uses compactly supported cohomology, even when q + is nonproper. For a nonproper q + one must not replace a q + * stalk by compactly supported cohomology without changing the functor. Neither variant is an extraordinarycostalkcalculation; that test is formulated for the composed operator in Appendix C.7.
Proof. 
Form the Cartesian square defined by the geometric point i y . Proper base change identifies i y * q + * with the direct image from F y . Apply that transformation to each of the three coefficient objects and to the map of coefficients induced by ϵ − . Naturality of proper base change gives the claimed identification of the connecting morphism. The last assertion uses the fact that a functor sends a zero map to zero. For q + ! use the base-change transformation for proper-support pushforward and the definition of compact-support cohomology. □

5.4. Which Hecke Comparison Is Actually Necessary?

The construction of (127) requires no comparison with a second Hecke category. Two stronger claims must be separated. The first is identification of one specified geometric kernel with a kernel obtained from a spherical Hodge module. The second is a natural comparison of Hecke actions on two independently defined coefficient categories.
Let G = G L 2 , G O = G ( C [ [ w ] ] ) , and Gr G = G ( K ) / G O . Fedorov constructs a convolution tensor category MHM G O ( Gr G ) , whose objects are equivariant mixed Hodge modules; the mixed Satake statement is formulated in Theorem 3 and treated in Section 5 of [11]. This category is not MHM ( X ) for an arbitrary Higgs-modification space X.
Lemma 12
(An actual framed map to the minuscule Schubert variety). Fix the framing and the reference lattice Λ 0 . The lower-lattice moduli scheme is P ( Λ 0 / w Λ 0 ) , identified with the minuscule Schubert variety S μ ≃ P 1 in the appropriate component of Gr G L 2 . On the framed incidence chart there is a defined algebraic morphism
h μ : H μ fr ⟶ S μ , ( m , ℓ ) ⟼ ρ − 1 ( ℓ ) ,
where the last expression means the lattice corresponding to ℓ under Proposition 31. For the ordinary unshifted unit coefficient 1 S μ = a S μ * Q H we have a canonical pullback identification h μ * 1 S μ = 1 H μ fr . For the perverse-normalized object on the smooth S μ , IC S μ H = 1 S μ [ 1 ] , this yields h μ * IC S μ H ≃ 1 H μ fr [ 1 ] . This does not determine the independently chosen Hecke kernel K μ , its Hodge–Satake normalization, or a comparison on an unframed quotient.
Proof. 
A framed modification records the colength-one sublattice, hence the universal line in Λ 0 / w Λ 0 . The classifying map to its projective-line moduli space is algebraic and pulls back under base change, so its composite with the minuscule-lattice identification gives h μ . The unit coefficient is the pullback of the constant Hodge object from the point; pseudofunctoriality of pullback gives h μ * a S μ * Q H = a H μ fr * Q H . Exactness of h μ * carries its cohomological shift [ 1 ] unchanged. No perverse t-exactness of this pullback is used. □
Corollary 5
(The unit-kernel Hecke operation is constructed). On the framed algebraic correspondence where q + exists, define
K μ unit = h μ * IC S μ H [ − 1 ] ≃ 1 H μ fr .
With this specified kernel the operator is the genuine derived functor
H μ unit ( M ) = q + * p − * M ,
whenever the stated boundedness conditions hold. It carries the chosen connecting arrow to q + * p − * ( ϵ − ) . Its existence makes no claim about identification with the independently normalized kernel used in [4].
Proof. 
Lemma 12 identifies the specified pullback kernel with the tensor unit. The tensor unit satisfies 1 ⊗ p − * M ≃ p − * M naturally. Applying q + * gives the asserted formula and its action on the specified morphism. Derived pullback and direct image are exact, regardless of perverse t-exactness. □
Definition 26
(Restricted spherical kernel comparison). Let S μ ⊂ Gr G be the minuscule Schubert variety, let S μ be a specified normalized equivariant Hodge module supported on S μ , and let h : H μ fr → S μ be an algebraic map compatible with the chosen framed modification. A restricted kernel comparison consists of a chosen pullback h † (either h * or h ! with specified shifts and Tate twists) and an isomorphism in the relevant derived MHM category
γ μ : h † S μ → ∼ K μ .
If the construction is to descend, the map and γ μ must also be compatible with the framing-change action.
Proposition 35
(Kernel identification implies operator identification). Assume the data of Definition 26 and a correspondence for which both resulting pull–tensor–push expressions are defined. Then γ μ induces a natural isomorphism of the two exact operators associated with the kernels h † S μ and K μ .
Proof. 
For each M ∈ D H ( X ) , tensor the isomorphism γ μ with p * M and apply q * . This gives
q * h † S μ ⊗ L p * M → ∼ q * K μ ⊗ L p * M .
Functoriality of tensor product and pushforward gives naturality in M. Each map is an isomorphism because it is the image of an isomorphism under an exact functor. □
Remark 39
(Limits of the restricted comparison). Neither an algebraic map h nor the isomorphism γ μ is inferred from the common P 1 parameter space. Even when both exist, Proposition 35 does not prove compatibility with a second convolution kernel. That requires an identification of the two-step geometric correspondence and the convolution of the chosen kernels. The diagrammatic mixed-perverse categories of [12] are distinct from Saito’s MHM category; their grading shifts are not automatically Hodge Tate twists. No diagrammatic-to-MHM functor is required here.
Proposition 36
(Localized one-step survival for the selected unit kernel). Let the framed source evaluation, admissible target map q + , and coefficient normalization agree with those of [4]. In particular, use the selected unit-kernel operation on the genuinely modified-Higgs correspondence, with the equivariance, framing-descent, and support hypotheses of those theorems. On the rank-one admissible target open U of the cited construction, the specified corrected attaching arrow has the nonzero star and proper-support images asserted there. On the finite saturated open U ′ ⊂ U , the higher degree-two derived attachment is nonzero in the sense and with the perverse normalization specified in the cited theorem.
These are assertions about theactual transported arrowon the selected target. They do not follow merely from the exactness of q + * p − * or from the auxiliary small-resolution calculation of Corollary 4.
Proof. 
The cited theorems supply nonzero target morphisms on their specified admissible opens, for their independently fixed kernel, target operation, support and descent conventions. The evaluated conifold arrow of Definition 24 is the source arrow used there, once the fixed local costalk identification, evaluation, coefficient kernel, and cohomological/Tate normalizations are matched. The claimed conclusions are therefore imports of the two cited geometric theorems. No vanishing or nonvanishing of a target arrow is inferred from the formal exactness of the pull–tensor–push operation. In particular q line and q + are not exchanged, and no claim for arbitrary kernels or unframed stacks is added. □
Remark 40
(The two distinct normalizations). Throughout Section 6 the proper-support operator uses q ! ( K ⊗ p * ( − ) ) , denoted here by H μ , ! * ; it is not the q ! p ! operator H μ , ! ! . The distinction concerns the actual pullback of the marked arrow, not merely a choice of notation. A comparison with exceptional pullback requires a purity or duality isomorphism on the selected correspondence, including all shifts and Tate twists. The imported one-step nonvanishing statements remain in their original star/shriek conventions.
Remark 41
(Relation to the second-step costalk criterion). The actual one-step source map e 0 and the target morphism q + are the data from which the composed evaluation t = e 0 ∘ p 02 and genuine target map q 02 must be built in Section 6. The smooth-purity criterion proved in Appendix C.7 requires smoothness of thiscompositenear the chosen target fiber, a nonempty proper fiber mapping to the node, and the stipulated constant composed kernel, including shifts and twists. Neither projectivity of p − nor the exceptional P 1 in Corollary 4 verifies those hypotheses. Even when the criterion applies, the perverse-degree boundary maps on the twice-modified target require separate evaluation.

5.5. Natural Intertwining and Its Effect on the Yoneda Arrow

For a cross-category comparison, let C − and C + be independently defined stable categories, equipped with exact functors
H μ cat : C − → C + , R H , − : C − → D H ( X ) , R H , + : C + → D H ( Y ) .
The latter functors are assumed to be constructed from actual geometric objects and maps; they are not defined by prescribing the image of an unknown attachment.
Definition 27
(Attachment-level intertwiner). An attachment-level comparison is a natural transformation of exact functors
Θ μ : R H , + ∘ H μ cat ⟹ H μ ∘ R H , − ,
compatible with suspension. It is an intertwining isomorphism if its components are equivalences on the specified admissible subcategory.
Proposition 37
(Naturality on the specified attachment). Suppose Definition 27 applies and let α : C → A [ 1 ] be an arrow in C − . Then the following square commutes in the homotopy category, with a coherent homotopy determined by the natural transformation:
R H , + H μ cat ( C ) → R H , + H μ cat ( α ) R H , + H μ cat ( A ) [ 1 ] Θ μ , C ↓ Θ μ , C Θ μ , A [ 1 ] ↓ Θ μ , A [ 1 ] H μ R H , − ( C ) → H μ R H , − ( α ) H μ R H , − ( A ) [ 1 ] .
If both endpoint components are isomorphisms and the source arrow has already been identified with the marked conifold attachment, the comparison identifies the resulting transported arrows.
Proof. 
Apply naturality of Θ μ to α . The two composites with common source R H , + H μ cat ( C ) and common target H μ R H , − ( A [ 1 ] ) are then identified. Compatibility with suspension identifies the latter target with H μ R H , − ( A ) [ 1 ] and the component at A [ 1 ] with Θ μ , A [ 1 ] . This is precisely (132). If the endpoint components are equivalences, conjugation by them identifies the two morphisms. Identification with the conifold attachment additionally uses the independently constructed endpoint maps and the comparison square of Section 3.6. □
Remark 42
(Soundness and domain of validity). Proposition 37 establishes soundness of agivennatural transformation; it does not construct Θ μ , establish that its components are isomorphisms, or enlarge its domain. Coverage of the corrected conifold attachment requires an independently specified object α and verified endpoint comparisons. The ordinary MHM transport of Proposition 33 has no need for such a transformation.

5.6. Framing Descent, Equivariance, and Higher Attachments

Let H be the algebraic group acting by changes of framing on the selected framed charts, and write a : H × X → X and pr 2 : H × X → X for the action and projection. For a mixed Hodge module M on X, an equivariant structure is an isomorphism ϑ : pr 2 * M → a * M satisfying the cocycle identity on H × H × X , with the appropriate normalization when the chosen equivariant formalism uses smooth shifted pullback.
Definition 28
(Equivariant attachment). An equivariant attachment on X consists of equivariant objects A , C ∈ D b MHM ( X ) and a morphism α : C → A [ 1 ] commuting with the two equivariance isomorphisms after pullback to H × X . This condition is imposed on the morphism, not merely on the isomorphism classes of its endpoints.
Proposition 38
(Conditional descent of the transported arrow). Assume that the framed Hecke correspondence is H-equivariant, that its kernel is equipped with compatible equivariant data, and that a chosen equivariant six-functor or quotient-stack MHM formalism supports the pull–tensor–push operation. Then H μ carries equivariant attachment morphisms to equivariant attachment morphisms. If the chosen formalism admits effective descent to the unframed quotient, the transported arrow descends with its specified coherence data.
Proof. 
Equivariance of the correspondence provides compatible action squares for p and q. Equivariance of the kernel identifies its pullbacks along the action and projection maps on the incidence space. Apply the exchange isomorphisms of the assumed equivariant six-functor formalism to obtain equivariant structures on p * M , on K μ ⊗ L p * M , and on their q * -images. The cocycle identity is preserved by functoriality of these exchange isomorphisms. Applying the same operations to an equivariant morphism gives a commuting equivariance square for its image. Effective descent, if available in the selected formalism, then supplies the corresponding morphism on the quotient. □
Remark 43
(Invariant line versus invariant attachment). The existence of an H-stable geometric line is weaker than H-equivariance of a chosen normalized extension generator. If a one-dimensional extension space carries a nontrivial character of a stabilizer, its nonzero vectors need not be invariant. The character and its action on the actual attaching morphism must be computed before asserting descent of that normalization.
Remark 44
(Localized survival and higher degree). The nonvanishing of the star and shriek images on the specified unframed open U, and the higher perverse-degree attachment on U ′ ⊂ U , are imported with their precise hypotheses from [4]. These statements concernonemodification and do not imply nonvanishing after the second modification; see the actual target-costalk criterion in Appendix C.7. If the target arrow has the form R ′ → I ′ [ 2 ] , its degree-two shift is part of the derived datum; replacing it by an ordinary Ext 1 in the perverse heart would change the assertion. The comparison with two consecutive Hecke modifications is treated separately in Section 6.

6. Two-step Minuscule Convolution and Coherence

We now calculate two successive lower minuscule modifications, first on lattices and then on the Higgs-preserving subscheme. The point is to identify the actual composed correspondence before applying a coefficient functor: the intermediate lattice is retained, and its transformed Higgs field carries first-jet data absent from the original residue. We compute the scheme of second invariant lines and its zero-, one-, and two-point regimes, including the doubled-point degeneration; we then compute the corresponding rational mixed-Hodge direct images of the simplest coefficient. These calculations provide an explicit geometric test, not only a list of formal conditions.
For a pair of finite-type MHM correspondences and specified kernels, we construct the base-change transformation from adjunction units and counits, apply the projection formula, and derive the two-step composition equivalence on objects and on the marked attaching morphism. A proper-support variant provides a composition statement when intermediate target maps are not proper. The perverse-cohomology long exact sequence defines the actual boundary maps whose nonvanishing would detect the transported attachment; the separate extraordinary-costalk criterion can detect the derived arrow without determining those maps. An octahedral calculation keeps separate the defects of any two successive coefficient comparisons. Identification with spherical Satake projectors, or with filtered Rees or KKPY F-bundle realizations, remains a different map-construction problem.
The first-jet calculation below also determines the entire reduced punctual incidence over a fixed jet with zero original residue; it is not only a calculation after fixing the first modification. A second result proves that the attaching arrow becomes zero after strict restriction to the punctual source. We complement that negative stalk calculation with an extraordinary-costalk formula for the genuine composed target and a sufficient smooth-purity criterion for a nonzero transported arrow. The latter applies only after checking the actual evaluation composite and the composed coefficient kernel; it does not determine the numerical perverse boundary maps on the twice-modified Higgs target.
The construction has two complementary geometric levels. At the level of lattices, two elementary modifications are parametrized by a projective surface with an exceptional fiber. At the level of Higgs bundles, each intermediate lattice must be preserved by the corresponding transformed Higgs field; these equations can reduce the lattice fibers and depend on first-order Higgs data. We calculate both levels before invoking any categorical convolution statement. The pure Schubert convolution of [3], the one-step conifold Hecke transport of [4], and the Hodge–Satake construction of [11] provide related but nonidentical coefficient settings.
In the terminology of Definition 15, the arrow being transported is the attaching morphism of a marked singular conifold Hodge atom, after the independently specified evaluation pullback. This terminology does not change the source category, the correspondence kernel, or the hypotheses required for nonvanishing.
The principal results are the restricted derived-MHM composition formula, the complete scheme-theoretic punctual first-jet calculation, and the smooth-purity detection criterion for the actual transported attachment. The first two require the specified algebraic charts; the last requires additional verifiable geometric hypotheses, including smoothness of the full evaluation composite near the target fiber. A comparison with a second Hecke category, transport of Satake idempotents to Higgs kernels, and compatibility with Rees specialization require independently constructed maps and are not used to prove the internal composition formula.

6.1. Dependencies and Permissible Coefficients for Two Steps

Let C be a smooth complex algebraic curve, x ∈ C a closed point, and w a local parameter at x. Set
O = C [ [ w ] ] , K = C ( ( w ) ) , Λ 0 = O 2 .
All lattice calculations in this section are on the completed formal disc at x. Their application to the algebraic Hecke correspondence requires the algebraic framed charts and the descent hypotheses of [4].
We use D H ( X ) for a stable enhancement of D b MHM ( X ) on a separated finite-type complex scheme X, and h D H ( X ) for its homotopy category. The six operations, tensor product, proper base change, and projection formula are taken with the conventions of Section 2 and Section 5; their enhanced compatibility is available in the setting of [6,10].
The projective G L 2 convolution calculation of [3] takes place in a category of pure Hodge modules on Schubert targets. In its stated normalization the resolution F 2 → Gr ¯ ( 2 , 0 ) has an exceptional P 1 , and the corresponding proper image has a punctual Q H ( − 1 ) constituent. This is not the Q H ( − 2 ) quotient in the corrected conifold extension (14); the bases, coefficients, and extension maps are different. Fedorov constructs spherical convolution in equivariant Hodge modules on the affine Grassmannian, proving the pure statement and outlining the mixed extension [11]. Neither construction alone yields convolution on Higgs-incidence coefficients. The present finite-type MHM calculation uses only Saito–Tubach operations until a further kernel comparison is explicitly established.
Definition 29
(Admissible two-step coefficient datum). An admissible datum consists of separated finite-type complex schemes X 0 , X 1 , X 2 , Z 01 , Z 12 , morphisms
X 0 ← p 01 Z 01 → q 01 X 1 , X 1 ← p 12 Z 12 → q 12 X 2 ,
and kernels K 01 ∈ D H ( Z 01 ) and K 12 ∈ D H ( Z 12 ) , including their specified cohomological shifts and Tate twists. We require the indicated pullbacks, tensor products, and pushforwards to preserve the bounded constructible subcategories in use. For the composition theorem below we further assume that the square of Definition 31 satisfies the proper Beck–Chevalley isomorphism and the projection formula for these objects. A sufficient hypothesis for the former is properness of q 01 .
Remark 45
(Star and shriek). The datum fixes star direct images. When a map is not proper, the operation with q ! is a different functor. The auxiliary q line , * and q line , ! of [4] must not be identified with the genuine modified-Higgs target operation. Results for any of these functors transfer to another only through a constructed comparison.

6.2. The Convolution Correspondence

Definition 30
(A two-step lower minuscule chain). A two-step chain of type ( 1 , 0 ) + ( 1 , 0 ) is a sequence of O -lattices
Λ 0 ⊃ Λ 1 ⊃ Λ 2 , dim C ( Λ i − 1 / Λ i ) = 1 ( i = 1 , 2 ) ,
where each quotient is annihilated by w. Equivalently, w Λ i − 1 ⊂ Λ i ⊂ Λ i − 1 . The chain parameter space on the framed disc is denoted Gr ˜ μ , μ , where μ = ( 1 , 0 ) ; the map retaining only the terminal lattice is denoted
m : Gr ˜ μ , μ ⟶ Gr ¯ ( 2 , 0 ) .
The target denotes the closure of the relative-position- ( 2 , 0 ) stratum in the component of colength two.
Lemma 13
(Finite-jet dependence of two-step Higgs incidence). For a chain of Definition 30, let A ∈ Mat 2 ( O ) . Whether A Λ i ⊆ Λ i for i = 1 , 2 depends only on A mod w 2 . Consequently the space of two-step Higgs-compatible chains for matrices varying in B 2 : = Mat 2 ( O / ( w 2 ) ) is an algebraic closed incidence locus in B 2 × Gr ˜ μ , μ .
Proof. 
The quotient Λ 0 / Λ 2 has length two and is annihilated by w 2 by the elementary-divisor classification for the chain. Thus w 2 Λ 0 ⊆ Λ 2 ⊆ Λ 1 . If A − A ′ = w 2 B with B ∈ Mat 2 ( O ) , then ( A − A ′ ) Λ i ⊆ w 2 Λ 0 ⊆ Λ i for i = 1 , 2 , since Λ i ⊆ Λ 0 . Hence A Λ i ⊆ Λ i if and only if A ′ Λ i ⊆ Λ i for both i. On each affine chart of the lattice-chain parameter space, preservation of the two universal lattices is expressed by the vanishing of the maps to the finite locally free quotients after the appropriate local trivializations. These conditions are algebraic and closed. They descend to the asserted closed incidence subscheme. □
Remark 46
(What is—and is not—determined by two jets). Lemma 13 asserts only that theincidence conditionsfor the two lattices depend on A mod w 2 . It does not assert that the transformedfirst jetof the Higgs field is a function of that two-jet. After the first modification, the entry b ( w ) / w loses one power of w; computing it modulo w 2 requires b ( w ) modulo w 3 . Thus the finite two-jet incidence scheme can be identified with a suitable truncated flag scheme, while an actual iterated map between Higgs moduli charts must use the full field, or a sufficiently higher jet on each target. This distinction will be retained in the comparison with the algebraic two-step correspondence below.
Proposition 39
(Terminal relative positions and fibers). For a chain (133), the finite-length O -module Λ 0 / Λ 2 has exactly one of the isomorphism types
O / ( w 2 ) , ( O / ( w ) ) ⊕ 2 .
The first type has one intermediate lattice Λ 1 ; the second type has a P 1 of intermediate lattices. In particular, m has a point as its fiber over the relative-position ( 2 , 0 ) stratum and P 1 over the terminal lattice w Λ 0 of type ( 1 , 1 ) .
Proof. 
The module Q = Λ 0 / Λ 2 has length two over the discrete valuation ring O , hence the structure theorem for finite-length modules gives the two displayed possibilities. An intermediate lattice corresponds to a submodule Λ 1 / Λ 2 ⊂ Q of length one, with quotient also of length one. If Q ≃ O / ( w 2 ) , its unique submodule of length one is w Q . If Q ≃ ( O / ( w ) ) 2 , every one-dimensional C -subspace of Q is an O - submodule of length one; these are parametrized by P ( Q ) ≃ P 1 . For the latter type, w Λ 0 ⊂ Λ 2 and equality follows by comparing colengths. The stated fibers follow. □
Proposition 40
(The two-step lattice surface). The variety Gr ˜ μ , μ of framed lower-minuscule chains based at Λ 0 is a smooth projective surface isomorphic to the Hirzebruch surface F 2 . The terminal-lattice map m of Definition 30 is proper; its exceptional fiber is the P 1 of Proposition 39.
Proof. 
The first lattice Λ 1 is the inverse image of a line in Λ 0 / w Λ 0 ≃ C 2 , hence the base of first modifications is B 1 = P 1 . Write L ≃ O B 1 ( − 1 ) for the tautological line and Q = ( O B 1 ⊕ 2 ) / L ≃ O B 1 ( 1 ) for the tautological quotient. The second modification is a line in the rank-two vector bundle E : = Λ 1 / w Λ 1 on B 1 . The inclusions w Λ 0 ⊆ Λ 1 ⊆ Λ 0 give a short exact sequence
0 ⟶ w Λ 0 / w Λ 1 ⟶ E ⟶ Λ 1 / w Λ 0 ⟶ 0 .
Multiplication by w identifies the first term with Q, whereas the last term is L. Thus
0 ⟶ O B 1 ( 1 ) ⟶ E ⟶ O B 1 ( − 1 ) ⟶ 0 .
Its extension class belongs to Ext B 1 1 ( O ( − 1 ) , O ( 1 ) ) = H 1 ( B 1 , O ( 2 ) ) = 0 . Consequently E ≃ O ( 1 ) ⊕ O ( − 1 ) , and the second-line parameter space is P B 1 ( E ) ≃ F 2 . The surface is smooth and projective. The terminal-lattice map is a morphism to the projective Schubert closure, hence proper. Its fibers have already been calculated in Proposition 39. □
For Higgs modifications, the lattice chain carries more data than its terminal relative position. In a trivialization of a rank-two bundle near x, write a local Higgs field as φ = A ( w ) d w with
A ( w ) = a ( w ) b ( w ) c ( w ) d ( w ) ∈ Mat 2 ( O ) .
For the chosen holomorphic Higgs convention, a lattice Λ is admissible if A ( w ) Λ ⊆ Λ . Other pole or residue conventions require an altered condition and are not included silently.
Lemma 14
(Two successive Higgs-invariance equations). Assume that the first modification is the lattice Λ 1 = w O e 1 ⊕ O e 2 . Then A ( w ) Λ 1 ⊆ Λ 1 if and only if b ( w ) ∈ w O . Writing b ( w ) = w β ( w ) , the matrix of the induced endomorphism of Λ 1 in the basis ( w e 1 , e 2 ) is
A 1 ( w ) = a ( w ) β ( w ) w c ( w ) d ( w ) .
If the second modification is specified by a line ℓ 2 = [ u : v ] ⊂ Λ 1 / w Λ 1 , it is Higgs-admissible precisely when
( w c ) u 2 + ( d − a ) u v − β v 2 ≡ 0 ( mod w ) .
Here the entries in (135) are evaluated as elements of O / ( w ) .
Proof. 
The image of w e 1 under A equals w a e 1 + w c e 2 , which belongs to Λ 1 . The image of e 2 equals b e 1 + d e 2 ; it belongs to Λ 1 exactly when b ∈ w O . Under this condition, substitution into the basis ( w e 1 , e 2 ) gives (134). A line represented by ( u , v ) over O / ( w ) is invariant under the reduction of A 1 precisely when ( u , v ) and A 1 ( u , v ) are linearly dependent. The determinant of these two column vectors is w c u 2 + ( d − a ) u v − β v 2 , giving (135). A change of the adapted basis transports the equation by the corresponding projective coordinate change, so the incidence condition is intrinsic. □
Proposition 41
(Scheme-theoretic second-line classification). Fix Λ 1 = w O e 1 ⊕ O e 2 and let A ( w ) preserve it. Write
δ = d ( 0 ) − a ( 0 ) , β 0 = ( b / w ) ( 0 ) .
Over the chosen first lattice, the scheme of possible second lines in P [ u : v ] 1 is
D δ , β 0 : = V v ( δ u − β 0 v ) ⊂ P 1 .
It has the following three scheme-theoretic regimes:
(i)
If δ ≠ 0 , it is the disjoint union of two reduced points [ 1 : 0 ] and [ β 0 : δ ] .
(ii)
If δ = 0 and β 0 ≠ 0 , it is the length-two doubled point V ( v 2 ) supported on [ 1 : 0 ] .
(iii)
If δ = β 0 = 0 , its defining equation is zero and the entire P 1 is allowed.
The family of schemes over the parameter plane Spec C [ δ , β 0 ] is not flat at the origin.
Proof. 
Reducing (135) modulo w gives ( d ( 0 ) − a ( 0 ) ) u v − ( b / w ) ( 0 ) v 2 = 0 , since w c ≡ 0 ( mod w ) . This is exactly (136). If δ ≠ 0 , the factors v and δ u − β 0 v vanish at distinct points of P 1 . If δ = 0 and β 0 ≠ 0 , the homogeneous equation becomes − β 0 v 2 = 0 , the usual double structure of length two on the point v = 0 . If both coefficients vanish there is no equation. To see nonflatness, restrict to the line δ = 0 , β 0 = t , and the chart u = 1 . The coordinate ring is C [ t , v ] / ( t v 2 ) . The nonzero element v 2 is killed by t; the family is not flat over C [ t ] , and therefore cannot be flat on a neighborhood of the origin in the parameter plane. □
Proposition 42
(Cohomology of the three explicit second-line fibers). Let a δ , β 0 : D δ , β 0 → pt be the structure map. For this calculation, take the reduction D δ , β 0 red , and use its derived constant rational Hodge coefficient without an additional perverse shift. Then the cohomology objects of R Γ ( D δ , β 0 red , 1 D δ , β 0 red ) are respectively:
( δ , β 0 ) H 0 H 2 δ ≠ 0 Q H ⊕ Q H 0 δ = 0 , β 0 ≠ 0 Q H 0 δ = β 0 = 0 Q H Q H ( − 1 )
with all other cohomology objects zero. Thus even the unshifted constant MHM coefficient exhibits distinct direct images in these three regimes. It detects the change from a full P 1 to a point, but not the nilpotent multiplicity of the doubled point.
Proof. 
By Proposition 41, the reductions of the three fibers are two points, one point, and P 1 , respectively. Constructible sheaf/Hodge-module cohomology of the unshifted constant coefficient on a nilpotent thickening agrees with that of its reduction. A finite discrete space with r points has H 0 ≃ ( Q H ) ⊕ r and no higher cohomology. For P 1 the unit and fundamental class give H 0 = Q H and H 2 = Q H ( − 1 ) ; all other groups vanish. This proves the table. It concerns the coefficient objects after pushforward, not the image of the conifold connecting morphism, which must be evaluated separately. □

The whole reduced punctual first-jet incidence

The preceding proposition holds after a first lattice has been fixed. For the complete punctual fiber we retain all first lower lattices simultaneously. Fix a matrix first jet
A ( w ) = w B + O ( w 2 ) , B = a b c d , A ( 0 ) = 0 .
Write π : F 2 = P P 1 ( E ) → P 1 for the first-line projection of Proposition 40. Put L = O P 1 ( − 1 ) and Q = O P 1 ( 1 ) , and let S = P ( Q ) ⊂ F 2 be the distinguished section corresponding to the second lattice w Λ 0 . The induced off-diagonal coefficient of B is the section
β B : L ⟶ Q , β B ∈ H 0 ( P 1 , O ( 2 ) ) , β B ( [ s : t ] ) ∼ c s 2 + ( d − a ) s t − b t 2 .
The indicated polynomial is a local homogeneous expression for β B and not an independently chosen scalar normalization.
Theorem 4
(Entire punctual first-jet incidence). Let Z B ⊂ F 2 be the closed scheme of chains preserving the fixed first jet (137). In adapted local coordinates [ u : v ] on the second-line fiber, where v = 0 cuts out S, its equation is
− β B ( ℓ ) v 2 = 0 .
These local equations are compatible under changes of adapted frame and therefore define the global incidence scheme. If β B ¬ ≡ 0 , the section S is generically doubled: on the open set β B ( ℓ ) ≠ 0 the defining ideal is ( v 2 ) , whereas the reduced section is defined by ( v ) . This nonreduced structure is not detected by the ordinary constant constructible coefficient. If B is scalar, the equation vanishes identically and Z B = F 2 is reduced; thus the multiplicity-two assertion does not include the scalar case. Write Z B = ( Z B ) red . Then:
  • If B has two distinct eigenlines ℓ 1 , ℓ 2 , then Z B = S ∪ π − 1 ( ℓ 1 ) ∪ π − 1 ( ℓ 2 ) , a connected tree of three smooth rational curves.
  • If B is nonscalar with one eigenline ℓ 0 , then Z B = S ∪ π − 1 ( ℓ 0 ) , two smooth rational curves meeting in a point. The eigenline zero divisor of β B is nonreduced before reduction.
  • If B is scalar, β B = 0 and Z B = F 2 .
For the reduced constant rational coefficient, the nonzero cohomology groups, with their rational mixed Hodge structures, are
B H 0 ( Z B ) H 2 ( Z B ) H 4 ( Z B ) two eigenlines Q H Q H ( − 1 ) ⊕ 3 0 one Jordan eigenline Q H Q H ( − 1 ) ⊕ 2 0 scalar Q H Q H ( − 1 ) ⊕ 2 Q H ( − 2 )
and all odd cohomology vanishes. These are ordinary cohomology objects, not numerical attaching-boundary maps.
Proof. 
Every first lower lattice is invariant under A ( 0 ) = 0 . Choosing an adapted frame for the first line ℓ and substituting the first jet into (134) gives the reduction 0 β B ( ℓ ) 0 0 . The invariant-second-line determinant is therefore − β B ( ℓ ) v 2 , proving the scheme equation. Away from V ( β B ) the reduced second-line fiber consists only of the section point v = 0 ; over V ( β B ) the entire fiber is allowed. The zero divisor of the degree-two section has two distinct points, one doubled point, or is all of P 1 , according as B has two eigenlines, one Jordan eigenline, or is scalar. The vertical fibers at distinct base points are disjoint and each intersects S in one point. Mayer–Vietoris for these trees of rational curves gives H 0 = Q H , one copy of Q H ( − 1 ) per irreducible rational curve in H 2 , and no odd cohomology. For the scalar jet the entire surface is F 2 , with its standard cell decomposition and H 4 = Q H ( − 2 ) . Passing to the reduction is appropriate for the rational constructible coefficient. In contrast, the coherent structure sheaf remembers the square-zero normal direction along the generic doubled section; this observation does not imply that every coherent or Rees coefficient detects it. □
Proposition 43
(Terminal-lattice direct image on the reduced punctual incidence). Let m B : Z B → C B = m ( Z B ) be the restriction of the terminal-lattice map, let o B = m ( S ) , and use ordinary derived constant rational coefficients on Z B . The fibers of m B are points except for S ≃ P 1 over o B . Consequently, in theordinary cohomology-sheaft-structure,
R 0 ( m B ) * Q Z B ≅ Q C B , R 2 ( m B ) * Q Z B ≅ Q { o B } ( − 1 ) , R q ( m B ) * Q Z B = 0 ( q ≠ 0 , 2 ) .
If one independently chooses to tensor the constant coefficient with the pullback of Q H ( − 2 ) from the point, the projection formula applies anexternalTate twist to every direct-image term: the degree-zero coefficient becomes Q H ( − 2 ) and the degree-two punctual coefficient becomes Q H ( − 3 ) . The untwisted formulas above concern the constant coefficient and do not by themselves determine the kernel or its cohomological shift in the genuine Higgs correspondence. No morphism from the conifold attaching triangle has yet been evaluated.
Proof. 
For the ordinary terminal-lattice map, the length-two quotient Λ 0 / Λ 2 is either cyclic of length two, with a unique intermediate lattice, or is C 2 annihilated by w, with a projective line of intermediate lattices. The latter occurs exactly over the terminal lattice w Λ 0 , and its whole intermediate-line fiber is the section S. The reduced punctual incidence contains S for every jet B. Away from that section the restricted terminal map is one-to-one. Proper base change computes the direct-image stalks from point or P 1 fibers: H 0 = Q in either case and H 2 = Q ( − 1 ) only at o B . The resulting degree-zero sheaf is constant on the connected image; the claimed higher skyscraper and vanishing follow. All Hodge twists refer to the rational Hodge realization of these constant-coefficient direct images. □
Example 2
(First-order dependence of the second line). For the first lattice Λ 1 = w O e 1 ⊕ O e 2 , compare
A ( 0 ) ( w ) = 0 0 0 0 , A ( 1 ) ( w ) = 0 w 0 0 .
Both induce the zero endomorphism on Λ 0 / w Λ 0 , and both preserve Λ 1 . For A ( 0 ) the second modification is parametrized by the complete P 1 . For A ( 1 ) , formula (134) gives A 1 = 0 1 0 0 . The invariant-line equation is − v 2 = 0 on [ u : v ] ∈ P 1 . Its underlying reduced support is the single line [ 1 : 0 ] ; scheme-theoretically the locus is the nonreduced length-two divisor defined by v 2 . Thus identical zeroth-order Higgs data can produce different second-step incidence schemes.
Proposition 44
(Properness of the bounded Higgs-incidence model). Put
B 2 = Mat 2 ( O / ( w 2 ) ) .
Write Z μ , μ Higgs for the closed incidence subscheme of B 2 × Gr ˜ μ , μ from Lemma 13. The morphism
Z μ , μ Higgs ⟶ B 2 × Gr ¯ ( 2 , 0 )
recording the matrix jet and terminal lattice is proper. Its fiber at a fixed jet and terminal lattice is the closed subscheme of the corresponding lattice fiber on which both Higgs-invariance conditions hold.
Proof. 
By Proposition 40, the map m is projective. Hence id B 2 × m is projective and in particular proper. Its restriction to the closed subscheme Z μ , μ Higgs is proper. The assertion about fibers follows by base change of the defining incidence equations. No assertion about the dimensions, smoothness, or reducedness of these restricted fibers follows; Example 2 exhibits nonreduced fibers. □
Remark 47
(Geometry retained by the calculation). The first Hecke line satisfies an invariance equation for the original matrix, while the second satisfies (135) for the modified matrix. Thus the Higgs-compatible two-step incidence is a closed incidence locus in the ordinary lattice-chain space, not generally the full convolution variety. Proposition 39 computes fibers of theunconstrainedlattice map; its P 1 fiber must not be assigned to a Higgs fiber without imposing both equations. No smoothness or Tor-independence of the Higgs-incidence locus follows from the lattice calculation.
Proposition 45
(The truncated two-step correspondence as a fiber product). Let B 2 = Mat 2 ( O / ( w 2 ) ) and let Gr ˜ μ , μ be the chain scheme. The closed locus Z μ , μ Higgs represents the following functor on finite-type C -schemes S:
( A 2 , Λ 1 , S , Λ 2 , S ) : A 2 ∈ B 2 ( S ) , Λ 0 , S ⊃ Λ 1 , S ⊃ Λ 2 , S , Λ i − 1 , S / Λ i , S locally free of rank one , A 2 Λ 1 , S ⊆ Λ 1 , S , A 2 Λ 2 , S ⊆ Λ 2 , S .
The two inclusion conditions are the successive invariant-line conditions: the first is imposed on the original matrix and the second on its matrix in the modified lattice Λ 1 , S . Thus on the framed truncated model this scheme is exactly the iterated incidence functor of Higgs-preserving lattice flags. Identification with the global fiber product Z 01 × X 1 Z 12 of actual Higgs-moduli correspondences requires the chosen algebraic framed charts to represent these same successive modifications, including the full transformed Higgs field and their descent conditions.
Proof. 
The chain scheme represents pairs of universal rank-one quotient lattices. Given a point of B 2 × Gr ˜ μ , μ over S, the obstructions to preservation are the composed maps
Λ i , S → A 2 Λ 0 , S ⟶ Λ 0 , S / Λ i , S , i = 1 , 2 .
They vanish precisely when the two lattices are preserved. Their vanishing defines closed subschemes (in local universal frames, the entries are polynomial equations). By Yoneda, their common vanishing locus represents the stated functor. The first equation produces a well-defined endomorphism of Λ 1 , S ; the second asserts that this induced endomorphism preserves Λ 2 , S , equivalently its line in Λ 1 , S / w Λ 1 , S . In the adapted frame this is exactly (135). The last assertion is not a consequence of this finite-jet calculation: it requires an independent comparison with the moduli functors of the chosen algebraic charts. □
Definition 31
(Composed algebraic correspondence). For the datum of Definition 29, set
Z 02 : = Z 01 × X 1 Z 12 ,
with projections a : Z 02 → Z 01 and b : Z 02 → Z 12 , so that q 01 ∘ a = p 12 ∘ b . Set
p 02 : = p 01 ∘ a : Z 02 → X 0 , q 02 : = q 12 ∘ b : Z 02 → X 2 .
The composed kernel on Z 02 is
K 02 : = a * K 01 ⊗ b * K 12 .
The tensor product and pullbacks are derived in D H ( Z 02 ) . If a derived correspondence model is used instead, its relation to this classical fiber product and the coefficient functors must first be supplied; it is not implicit in this definition.

6.3. The Finite Two-Step Calculation before a General Action

For i = 01 , 12 , let
H i ( M ) = q i * K i ⊗ p i * M .
These are exact functors on the selected enhanced MHM categories under Definition 29. Define
H 02 ( M ) = q 02 * K 02 ⊗ p 02 * M .
Lemma 15
(Constructing the Beck–Chevalley morphism). For the Cartesian square of Definition 31, the base-change transformation
BC : p 12 * q 01 * ⟶ b * a *
is the following composite of the unit and counit of the adjunctions ( b * , b * ) and ( q 01 * , q 01 * ) :
p 12 * q 01 * F → η b b * b * p 12 * q 01 * F → ≃ b * a * q 01 * q 01 * F → b * a * ( ε q 01 ) b * a * F .
If q 01 is proper, the proper-base-change theorem makes (144) an equivalence on the selected derived MHM categories. Without that hypothesis it is a defined transformation, not automatically an equivalence. For separated finite-type maps, the proper-support exchange p 12 * q 01 ! → ∼ b ! a * is available in the six-functor formalism.
Proof. 
The Cartesian identity q 01 a = p 12 b gives the pseudofunctorial isomorphism b * p 12 * ≃ a * q 01 * . Insert it between the unit and counit displayed above. This gives the map on every coefficient object F, and naturality of the adjunction transformations gives a natural transformation. Its invertibility under properness is exactly the proper-base-change theorem, not a consequence of the existence of the mate. The proper-support variant is the base-change isomorphism for !-pushforward in the same six-operation formalism. □
Remark 48
(A completely specified test kernel for two steps). On any chosen pair of actual framed algebraic correspondences, put K 01 = 1 Z 01 and K 12 = 1 Z 12 . By the monoidal unit identities and pseudofunctoriality of pullback,
a * K 01 ⊗ b * K 12 ≃ 1 Z 02 .
The corresponding composed operation is the explicit functor q 02 * p 02 * (or q 02 ! p 02 * for proper-support transport). This is an actual coefficient choice with no Satake-to-Higgs kernel isomorphism assumed. Evaluating theconifold attaching arrowunder it is still a separate morphism computation.
Lemma 16
(The projection-formula morphism on the composed square). Let F ∈ D H ( Z 02 ) and G ∈ D H ( Z 12 ) . The canonical map used to move a kernel through b * is the composite
G ⊗ b * F → η b b * b * ( G ⊗ b * F ) → ≃ b * ( b * G ⊗ b * b * F ) → b * ( 1 ⊗ ε b ) b * ( b * G ⊗ F ) .
Under the projection-formula hypotheses this transformation is an isomorphism. In particular it applies to G = K 12 and F = a * ( K 01 ⊗ p 01 * M ) .
Proof. 
The first arrow is the unit of ( b * , b * ) , the second is the symmetric-monoidal pullback identification, and the third is induced by the counit b * b * F → F . These are natural in both variables. The projection-formula theorem states that the composite is invertible on the selected coefficient categories. Thus its use in the convolution calculation is the application of a defined transformation, not the assertion that two unrelated tensor products are equal. □
Proposition 46
(Restricted two-step kernel composition). Suppose that the datum of Definition 29 satisfies proper base change for the Cartesian square defining Z 02 and the projection formula for the displayed kernels. Then there is a natural equivalence of exact functors
κ H : H 12 ∘ H 01 → ∼ H 02 .
The equivalence preserves the specified cohomological shifts and Tate twists already incorporated into the kernels.
Proof. 
For M ∈ D H ( X 0 ) , expand the left-hand side:
H 12 H 01 ( M ) = q 12 * K 12 ⊗ p 12 * q 01 * ( K 01 ⊗ p 01 * M ) .
The square
Z 02 → a Z 01 b ↓ b q 01 ↓ q 01 Z 12 → p 12 X 1
is Cartesian. The assumed Beck–Chevalley map is an isomorphism
p 12 * q 01 * → ∼ b * a * .
Substitute it into the preceding expression and use the projection formula for b:
H 12 H 01 ( M ) ≃ q 12 * K 12 ⊗ b * a * ( K 01 ⊗ p 01 * M ) ≃ q 12 * b * b * K 12 ⊗ a * K 01 ⊗ a * p 01 * M ≃ ( q 12 b ) * K 02 ⊗ ( p 01 a ) * M = H 02 ( M ) .
The third line uses pseudofunctoriality of pushforward, pullback, and the symmetry of tensor product. Every step is natural in M; therefore these maps assemble into the stated equivalence of exact functors. Since the shifts and Tate twists are part of the input kernels, all identifications respect their chosen normalizations. □
Proposition 47
(Two-step proper-support composition). For the finite-type algebraic datum of Definition 29, define a second pair of functors with the same pullback and tensor kernels but !-pushforward:
H i j c ( M ) : = q i j ! ( K i j ⊗ p i j * M ) , i j = 01 , 12 , 02 .
Assume these expressions preserve the bounded coefficient subcategories in use and the projection formulas apply. Then, without requiring properness of q 01 , the proper-support base-change isomorphism determines a natural equivalence
κ c : H 12 c H 01 c → ∼ H 02 c .
It acts on every specified attaching arrow by the same natural comparison square as (149). This q ! p * operator is not the distinct q ! p ! normalization of Definition 25.
Proof. 
Apply the proper-support base-change transformation of Lemma 15 to obtain
H 12 c H 01 c ( M ) = q 12 ! K 12 ⊗ p 12 * q 01 ! ( K 01 ⊗ p 01 * M ) ≃ q 12 ! K 12 ⊗ b ! a * ( K 01 ⊗ p 01 * M ) ≃ q 12 ! b ! b * K 12 ⊗ a * K 01 ⊗ ( p 01 a ) * M ≃ q 02 ! ( K 02 ⊗ p 02 * M ) .
The second equivalence uses the projection formula for b ! ; the last uses pseudofunctorial composition of !-pushforward. Every displayed transformation is natural in M and respects suspension. The final assertion follows by applying this naturality to the specified connecting arrow. Notice that no properness is claimed for the actual Higgs target maps. □
Lemma 17
(Naturality of the base-change and kernel maps on the arrow). Put E R = K 01 ⊗ p 01 * R − H , E I = K 01 ⊗ p 01 * I − H , and let e = 1 K 01 ⊗ p 01 * ϵ − : E R → E I [ 1 ] . The actual Beck–Chevalley morphism of Lemma 15 gives a commutative square of derived MHM objects
p 12 * q 01 * E R → p 12 * q 01 * e p 12 * q 01 * E I [ 1 ] BC E R ↓ BC E R BC E I [ 1 ] ↓ BC E I [ 1 ] b * a * E R → b * a * e b * a * E I [ 1 ] .
Tensoring this square by K 12 and applying the specific projection-formula map (145) constructs the comparison between the two-step connecting arrows. Under the proper-base-change and projection-formula hypotheses, this comparison is the isomorphism (149).
Proof. 
Expand the two paths in (148) using the unit η b and the counit ε q 01 as in (144). Naturality of these two adjunction transformations with respect to e identifies the paths. The monoidal pullback identification also commutes with e, so the intermediate square is commutative. Apply tensor with K 12 and then the transformation (145); its unit and counit are natural in F, hence preserve this commuting square. Finally apply q 12 * and compose pushforwards. Every map is constructed from the displayed geometric square, units, counits and tensor transformations. The isomorphism conditions follow from proper base change and the projection formula. □
Corollary 6
(Composition on the specified attachment). Let
I − H ⟶ P − H ⟶ R − H → ϵ − I − H [ 1 ]
be the distinguished triangle obtained by the specified evaluation pullback in Section 5. Under the hypotheses of Proposition 46, the square
H 12 H 01 ( R − H ) → H 12 H 01 ( ϵ − ) H 12 H 01 ( I − H ) [ 1 ] ∼ κ R − H H ↓ ∼ κ R − H H κ I − H H [ 1 ] ↓ κ I − H H [ 1 ] H 02 ( R − H ) → H 02 ( ϵ − ) H 02 ( I − H ) [ 1 ]
commutes in the homotopy category, with a specified coherent homotopy in the enhancement. No nonvanishing assertion follows solely from this square.
Proof. 
Apply the natural equivalence κ H to the morphism ϵ − : R − H → I − H [ 1 ] . Naturality gives
κ I − H [ 1 ] H ∘ H 12 H 01 ( ϵ − ) = H 02 ( ϵ − ) ∘ κ R − H H .
Exactness identifies κ I − H [ 1 ] H with κ I − H H [ 1 ] . This is (149). Coherence is inherited from the stable enhancement and the exchange transformations used in Proposition 46. An exact functor need not detect a nonzero arrow, which proves the final qualification. □
Proposition 48
(Strict punctual restriction kills the forward attachment). Let i o : { o } ↪ X det be the ordinary double point, I H = I C X det H , and R o H = i o * Q o H ( − 2 ) . For every arrow ϵ o : R o H → I H [ 1 ] in D b MHM ( X det ) , one has
i o * ϵ o = 0 in D b MHS Q .
If a restricted source morphism p : Z o → X det factors through i o , the pulled-back arrow p * ϵ o is zero. Its image under any defined tensor-with-kernel and q * or q ! transport is also zero, and consequently all the ordinary or perverse cohomology maps of thisstrictly punctualtransport vanish. No conclusion about the full two-step modified-Higgs pushforward follows from this assertion.
Proof. 
The ODP intersection-complex stalk computation with the fixed threefold normalization is
H k ( i o * I H ) = Q H , k = − 3 , Q H ( − 1 ) , k = − 1 , 0 , otherwise .
The only cohomology of i o * I H [ 1 ] lies in degrees − 4 and − 2 . Rational mixed Hodge structures have homological dimension one: Ext MHS Q r ( H 1 , H 2 ) = 0 for r ≥ 2 (cf. [24]). By the hyper-Ext spectral sequence (or the truncation triangles), a degree-zero morphism from Q o H ( − 2 ) to i o * I H [ 1 ] could only arise from Ext 4 or Ext 2 , and both vanish. Therefore
Hom D b MHS Q ( Q o H ( − 2 ) , i o * I H [ 1 ] ) = 0 .
Since i o * R o H ≃ Q o H ( − 2 ) , the image i o * ϵ o lies in this zero group. If p = i o ∘ p o , pseudofunctoriality gives p * ϵ o = p o * i o * ϵ o = 0 . Exact pullback, tensor with a fixed kernel, and derived direct image send the zero morphism to zero. All its cohomology maps are thus zero as well. This argument crucially invokes the homological dimension of rational MHS; it is not a consequence of perverse t-structure orthogonality alone. □
Remark 49
(Punctual restriction is not the genuine second Hecke image). The theorem applies when thesource map actually factorsthrough the closed point. The bounded jet-incidence fiber and its terminal-lattice map describe geometry over a fixed punctual source; they do not retain the transverse source directions used by the original variation-cone attachment. The full first-to-second Higgs correspondence, its target morphism q 12 , and its coefficient complex cannot be replaced by the punctual map m B . In particular the vanishing of i o * ϵ o does not imply H 02 ( ϵ − ) = 0 , nor does the cohomology in (141) prove nonvanishing.
Proposition 49
(The two-step perverse boundary is a calculable map). Put A 02 = H 02 ( I − H ) , P 02 = H 02 ( P − H ) and C 02 = H 02 ( R − H ) , and write ϵ 02 = H 02 ( ϵ − ) . For each n define
δ 02 n : = H n ( ϵ 02 ) : H n ( C 02 ) ⟶ H n + 1 ( A 02 )
in the abelian category MHM ( X 2 ) . Then the two-step attachment triangle gives
0 ⟶ coker ( δ 02 n − 1 ) ⟶ H n ( P 02 ) ⟶ ker ( δ 02 n ) ⟶ 0 .
Under the equivalence κ H , the map δ 02 n is identified with the corresponding cohomology map of H 12 H 01 ( ϵ − ) . In particular, its computation requires the actual coefficient complex on the two-step fiber product, and a nonzero δ 02 n proves ϵ 02 ≠ 0 .
Proof. 
Apply H n to the image under H 02 of the source triangle, obtaining the five-term exact segment
H n − 1 ( C 02 ) → δ 02 n − 1 H n ( A 02 ) ⟶ H n ( P 02 ) ⟶ H n ( C 02 ) → δ 02 n H n + 1 ( A 02 ) .
The image of the middle map is ker δ 02 n and its kernel is the image of H n ( A 02 ) , which is its quotient by im δ 02 n − 1 . This constructs (152). Since κ H is a natural equivalence of exact functors, applying it to ϵ − produces a commutative square. Passing to cohomology identifies the two corresponding δ -maps. The final detection statement is immediate: a zero derived morphism induces zero maps on all cohomology objects. □
Remark 50
(Derived nonvanishing and perverse boundary are different claims). The smooth-purity theorem below detects the derived arrow ϵ 02 without requiring any particular δ 02 n to be nonzero. The degreewise maps in (151) are the cohomology maps of that arrow, and a nonzero derived morphism need not be detected on the cohomology objects alone. Hence the precise remaining numerical task is to calculate these maps from the genuine second-target coefficient morphism; it is not a prerequisite to the conditional derived nonvanishing theorem.
Proposition 50
(Boundary of the numerical R6 computation). The morphisms δ 02 n of Proposition 49 are determined by theactualtwice-modified Higgs target map q 02 , the composed kernel K 02 , the evaluated source arrow, and the exchange morphisms used in the selected *- or !-normalization. The reduced punctual incidence of Theorem 4 and the endpoint images of Proposition 43 do not determine numerical values for these maps. In order to compute them, one must retain the source directions transverse to the conifold point, specify the genuine second target map and the resulting coefficient morphism, and evaluate its perverse costalk or Gysin image. Lemma 19 gives the extraordinary fiber costalk formula. Theorem 5 proves derived nonvanishing from the checkable smooth-purity and constant-kernel hypotheses, but those hypotheses must still be verified on the actual two-step Higgs charts. Projection to the Sym 2 and determinant channels additionally requires an independently verified comparison of the spherical convolution kernel with the Higgs kernel.
Proof. 
The formula defining δ 02 n takes cohomology of the specific derived morphism H 02 ( ϵ − ) . By contrast, (141) computes direct-image objects of a constant coefficient under a restricted map. Even a complete description of the latter objects cannot recover a map between them. More decisively, Theorem 48 shows the actual conifold arrow becomes zero under strict punctual pullback; its full-image value can therefore only be studied before those transverse directions are discarded. The Satake decomposition is an identity in a separate coefficient category and acts on the Higgs morphism only after a geometric kernel comparison is given. None of these missing maps is supplied by the punctual fiber table. □
Lemma 18
(Stalkwise fiber model for the two-step arrow). If q 02 is proper on the support of the composed kernel and coefficients, then for a geometric point y → X 2 and F y = q 02 − 1 ( y ) the image of i y * ϵ 02 under proper base change is the explicitly defined morphism
R Γ F y , K 02 | F y ⊗ p 02 , y * R − H → R Γ ( K 02 ⊗ p 02 , y * ϵ − ) R Γ F y , K 02 | F y ⊗ p 02 , y * I − H [ 1 ] .
It detects the global arrow if nonzero. This is a valid stalkwise formula, but in the strictly punctual regime it can vanish for the reason of Proposition 48; Lemma 19 supplies the separate extraordinary-costalk test, and Theorem 5 supplies a sufficient nonvanishing criterion under additional geometric hypotheses. Knowledge of thecohomology groupsof F y without the coefficient map p 02 , y * ϵ − does not determine its value. When q 02 is nonproper, the parallel statement uses q 02 ! and compactly supported cohomology.
Proof. 
Base-change the composed correspondence along y → X 2 ; apply proper base change to each endpoint and to K 02 ⊗ p 02 * ϵ − . Naturality identifies the resulting arrow with (153). If that map is nonzero, its antecedent could not have been zero. The last assertion uses the !-base-change theorem and compactly supported direct image. □
Lemma 19
(Extraordinary fiber model for the composed attachment). Retain the genuine composed correspondence of Definition 31 and the arrow ϵ 02 = H 02 ( ϵ − ) . Let i y : { y } ↪ X 2 be a geometric closed point, F y = Z 02 × X 2 { y } its scheme-theoretic fiber, k y : F y → Z 02 its inclusion, and a y : F y → { y } the structure morphism. Assume that q 02 is proper on the stated coefficient support and that extraordinary proper base change is valid in the selected derived mixed-Hodge-module categories. Put
M R = K 02 ⊗ L p 02 * R − H , M I = K 02 ⊗ L p 02 * I − H .
Then the costalk of theactualtwo-step arrow is the morphism
i y ! ϵ 02 : R a y * k y ! M R ⟶ R a y * k y ! M I [ 1 ] , induced by R a y * k y ! ( 1 K 02 ⊗ p 02 * ϵ − ) .
No tensor-exceptional-pullback interchange is implicit in this formula. If the morphism in (154) is nonzero, then ϵ 02 is nonzero. Its degreewise cohomology maps are the corresponding costalk detection tests, with no assertion about their values without a calculation.
Proof. 
The Cartesian square
F y → k y Z 02 a y ↓ a y q 02 ↓ q 02 { y } → i y X 2
induces the extraordinary proper-base-change isomorphism i y ! R q 02 * ≃ R a y * k y ! for the stated supported coefficients. Apply this natural isomorphism to both endpoints of M R → M I [ 1 ] and to the connecting map itself. Naturality produces (154). A functor sends the zero morphism to zero, so nonvanishing of its costalk image implies nonvanishing of ϵ 02 . □
Theorem 5
(Smooth-purity detection for the genuine two-step arrow). Retain the actual algebraic correspondence and the notation of Lemma 19. Let e : X 0 → X det be the specified source evaluation, ϵ − = e * ϵ o , and put t = e ∘ p 02 : Z 02 → X det . Suppose that q 02 is proper on the displayed coefficients, F y is nonempty, t is smooth of constant relative dimension r on an open neighborhood of F y , and the restricted map t ∘ k y : F y → X det factors through the node as t ∘ k y = i o ∘ a y . Assume that the actual composed kernel, including its coefficient structure, has a specified identification K 02 | W ≃ 1 W [ d ] ( m ) on a neighborhood W ⊃ F y , compatible with the transported arrow. If the ODP costalk coordinate satisfies H 0 ( i o ! ϵ o ) = c o id with c o ≠ 0 , then
i y ! ϵ 02 ≠ 0 , ϵ 02 ≠ 0 in D b MHM ( X 2 ) .
Here 1 W is the derived constant rational Hodge coefficient, with the shifts and Tate twists [ d ] ( m ) fixed as part of the actual kernel, and the nonvanishing is that of a derived morphism. No equality F y = t − 1 ( o ) , smoothness of F y , or separate smoothness of the correspondence space is imposed.
Proof. 
The unit-kernel identification permits the extraordinary fiber arrow of Lemma 19 to be written, up to the fixed common shift and twist [ d ] ( m ) , as R a y * k y ! t * ( ϵ o ) . Smooth purity for the morphism t identifies t ! ≃ t * ( r ) [ 2 r ] , or, equivalently, t * ≃ t ! ( − r ) [ − 2 r ] . Pseudofunctoriality of extraordinary pullback and the stipulated factorization t ∘ k y = i o ∘ a y therefore yield a canonical identification of the actual arrow
k y ! t * ϵ o [ d ] ( m ) ≃ k y ! t ! ϵ o [ d − 2 r ] ( m − r ) ≃ a y ! i o ! ϵ o [ d − 2 r ] ( m − r ) .
The displayed isomorphisms are natural in both endpoints and the given morphism. In particular, they do not interchange k y ! with tensor product by an arbitrary coefficient kernel.
Let ω F y : = a y ! 1 pt be the dualizing Hodge complex. The projection formula for the structure morphism to a point identifies the pushed-forward arrow, before its common shift and twist, with
R a y * a y ! ( i o ! ϵ o ) ≃ ( i o ! ϵ o ) ⊗ L R a y * ω F y .
Since F y is nonempty and proper, Verdier duality gives R a y * ω F y ≃ D pt R Γ ( F y , 1 F y ) . Its underlying rational complex has nonzero degree-zero cohomology, canonically dual to H 0 ( F y , Q ) . The source of i o ! ϵ o is the degree-zero punctual Tate object and the induced degree-zero cohomology map is c o id ≠ 0 . After rational realization the category of bounded complexes of Q -vector spaces is semisimple; tensoring this nonzero morphism with a nonzero complex has a nonzero degree-zero component on the tensor of those two cohomology groups. Hence the morphism in (157) is nonzero after rational realization, and therefore is nonzero in the derived mixed Hodge category. Restoring [ d − 2 r ] ( m − r ) and applying the extraordinary proper-base-change identification from Lemma 19 proves both claims. The detailed coefficient-level calculation is also given in Appendix C.7. □
Remark 51
(Geometric verification and the nonconstant-kernel boundary). The theorem gives sufficient geometric hypotheses; it does not claim that the actual twice-modified Higgs correspondence satisfies them. Smoothness of either e or p 02 alone does not imply that t = e ∘ p 02 is smooth near F y ; smoothness of both on compatible neighborhoods would suffice by stability under composition. Nor does the bounded properness of the truncated terminal-lattice map prove properness of the full target morphism q 02 . The requirement that t ( F y ) = { o } is independent of any equality between F y and t − 1 ( o ) . Indeed, a regular source at a point above the node isimpossibleunder the smoothness hypothesis on t. If z ∈ F y and t ( z ) = o , then smoothness of t at z together with regularity of Z 02 at z would imply regularity of X det at o by descent of regularity along smooth morphisms. Since o is singular, Z 02 is necessarily singular at every such point z. The theorem therefore requires smoothness of themapt near F y , not regularity of its source or fiber. For a nonconstant composed kernel one must instead construct the appropriate k y ! –tensor exchange and prove that its proper direct-image arrow is nonzero; smooth purity alone does not supply that kernel comparison. The punctual first-jet trees of Theorem 4 calculate constant-coefficient incidence geometry, not the kernel identification or smoothness hypotheses. In particular, the theorem does not compute any of the numerical perverse boundary maps δ 02 n or their Satake-channel projections. The canonical local coordinate is c o = − 1 ; for the corrected nonzero class c o = − λ o ≠ 0 , without a claim that λ o = 1 .
Remark 52
(Soundness and geometric coverage). Proposition 46 proves a composition formulafor every datum satisfying its hypotheses. Application to all framed Higgs-modification charts requires separate verification that their actual one- and two-step spaces, kernels, base-change maps, and descent satisfy those hypotheses. This coverage statement is not a consequence of formal naturality. The localized nonvanishing theorems of [4] likewise do not imply nonvanishing under a second Hecke functor.

6.4. The Composition-Coherence Diagram

Proposition 51
(An explicit octahedron for two coefficient comparisons). Fix a single algebraic correspondence X 0 ← p Z → q X 2 and three bounded complexes of MHM kernels on Z with composable chain maps K 0 → f K 1 → g K 2 . Use the cone convention of Definition 3. The chain maps
u : Cone ( f ) ⟶ Cone ( g f ) , u ( y , x ) = ( g y , x ) , v : Cone ( g f ) ⟶ Cone ( g ) , v ( z , x ) = ( z , f x )
fit into the octahedral distinguished triangle
Cone ( f ) → u Cone ( g f ) → v Cone ( g ) → ∂ oct Cone ( f ) [ 1 ] .
The connecting chain map is explicitly
∂ oct ( z , y ) = ( y , 0 ) , ( z , y ) ∈ K 2 n ⊕ K 1 n + 1 = Cone ( g ) n .
In particular the composite v u is chain-null-homotopic via h ( y , x ) = ( 0 , y ) , with
v u = d Cone ( g ) h + h d Cone ( f ) .
For each M ∈ D H ( X 0 ) , the exact functor q * ( − ⊗ p * M ) carries this triangle to an octahedron comparing the three correspondingcoefficient defects. Even if Cone ( g f ) becomes zero after that functor, this does not imply the two individual cones vanish.
Proof. 
In cohomological degree n we have Cone ( f ) n = K 1 n ⊕ K 0 n + 1 and Cone ( g ) n = K 2 n ⊕ K 1 n + 1 . The differential in Cone ( f ) is d f ( y , x ) = ( d K 1 y + f x , − d K 0 x ) , and similarly for g and g f . Direct computation gives
d g f u ( y , x ) = ( g d K 1 y + g f x , − d K 0 x ) = u d f ( y , x ) , d g v ( z , x ) = ( d K 2 z + g f x , − f d K 0 x ) = v d g f ( z , x ) .
Thus both maps in (158) are chain maps. The degree-minus-one map h : Cone ( f ) n → Cone ( g ) n − 1 sends ( y , x ) to ( 0 , y ) . Its differential identity is
( d g h + h d f ) ( y , x ) = ( g y , − d K 1 y ) + ( 0 , d K 1 y + f x ) = ( g y , f x ) = v u ( y , x ) .
The boundary map (160) is a chain map:
∂ oct d g ( z , y ) = ( − d K 1 y , 0 ) = d Cone ( f ) [ 1 ] ∂ oct ( z , y ) .
To check the triangle itself, form Cone ( u ) . Its cohomological degree-n term is
Cone ( u ) n = ( K 2 n ⊕ K 0 n + 1 ) ⊕ ( K 1 n + 1 ⊕ K 0 n + 2 ) .
The map
r : Cone ( u ) n ⟶ Cone ( g ) n , r ( z , x , y , x ′ ) = ( z , y + f x )
is a chain map: substituting the cone differentials gives d g r = r d Cone ( u ) . Its kernel consists of ( 0 , x , − f x , x ′ ) , and, with coordinates ( x , x ′ ) , has differential
( x , x ′ ) ⟼ ( − d K 0 x + x ′ , d K 0 x ′ ) .
The degree-minus-one operator ( x , x ′ ) ↦ ( 0 , x ) contracts the kernel: d h + h d = 1 . Hence r is a quasi-isomorphism. Under this identification, the standard map Cone ( g f ) → Cone ( u ) becomes v. The standard projection Cone ( u ) → Cone ( f ) [ 1 ] sends ( z , x , y , x ′ ) to ( y , x ′ ) ; its difference from ∂ oct r , which sends this element to ( y + f x , 0 ) , is the coboundary of the degree-minus-one homotopy ( z , x , y , x ′ ) ↦ ( 0 , − x ) . Consequently the connecting map is (160), and (159) is a distinguished triangle with the specified sign convention.
Applying q * ( − ⊗ p * M ) preserves this triangle, because tensor with a fixed coefficient and derived direct image are exact on the selected stable categories. A vanishing cone of the composite does not force either of the two separate cones to vanish, as follows directly from the distinction among the three vertices of the octahedron. □
We next distinguish the preceding internal MHM composition from a comparison with an independently constructed categorical Hecke operation. Let C 0 , C 1 , C 2 be stable categories, H 01 cat : C 0 → C 1 and H 12 cat : C 1 → C 2 exact functors, and R i : C i → D H ( X i ) exact comparison functors. These are additional data, not categories or functors defined by the conifold attachment. Assume given natural transformations
Θ 01 : R 1 H 01 cat ⇒ H 01 R 0 , Θ 12 : R 2 H 12 cat ⇒ H 12 R 1 .
Assume also exact composed functors H 02 cat and H 02 , an isomorphism κ cat : H 12 cat H 01 cat → ∼ H 02 cat , the isomorphism κ H of (146), and a natural transformation Θ 02 : R 2 H 02 cat ⇒ H 02 R 0 .
Definition 32
(Two-step intertwining coherence). The transformations above satisfy two-step coherence if, as transformations from R 2 H 12 cat H 01 cat to H 02 R 0 , one has
Θ 02 ∘ R 2 ( κ cat ) = κ H R 0 ∘ ( H 12 Θ 01 ) ∘ ( Θ 12 H 01 cat ) .
The compositions and whiskering are interpreted in the indicated stable functor categories.
Proposition 52
(Reduction to a checked kernel comparison). Suppose all the comparison functors and transformations above are obtained by functorially applying the same six-operation exchange maps to specified diagrams of geometric correspondences and kernel morphisms. If the resulting two composites of kernel morphisms on the two-step correspondence agree, then (162) holds on the admissible source subcategory. It consequently holds on each attaching arrow in that subcategory.
Proof. 
The correspondence calculus carries a map of coefficient kernels to a natural transformation between their associated pull–tensor– push functors. Under the hypothesis, the two sides of (162) are obtained by applying this construction to the two stipulated composites of kernel maps. Their equality gives equality of the resulting transformations. Evaluation at an object gives the component identity. Evaluation at a morphism α : C → A [ 1 ] gives its naturality square; exactness identifies the component at A [ 1 ] with the shift of the component at A. This proves the assertion. The hypothesis concerns independently defined kernel morphisms; no comparison morphism is selected by requiring (162) itself. □
Remark 53
(Logical status of F4). The existence of κ H does not construct C i , R i , or Θ i j . Accordingly, Proposition 52 is a conditional cross-category assertion. The internal MHM composition of Corollary 6 requires none of these additional data and does not assert arbitrary higher coherence for independently specified actions.

6.5. Geometric Satake and the Tensor-Square Channels

Let G = G L 2 and let G ∨ ≃ G L 2 denote its Langlands dual group. Write V for its standard two-dimensional representation over Q . In Rep Q ( G ∨ ) one has the canonical decomposition
V ⊗ V ≃ Sym 2 V ⊕ det V .
Let H G pure be Fedorov’s pure G ( O ) -equivariant Hodge–Satake category, and let S μ denote the appropriately normalized Hodge intersection complex of the minuscule Schubert variety. The corresponding tensor-square decomposition takes place within H G pure , with the parity and Tate normalizations prescribed by Hodge–Satake; it is not simply an unshifted isomorphism between arbitrary Hodge modules [11,25].
Remark 54
(Different punctual Tate terms). The punctual Q H ( − 1 ) summand in the proper-image calculation of [3] is a contribution to pure Schubert convolution. The forward conifold quotient R Σ H = ⨁ p ∈ Σ i p * Q H ( − 2 ) is part of a mixed extension on the singular threefold. Neither the Sym 2 V nor the det V summand is thereby identified with R Σ H .

6.6. Satake-to-Higgs Kernel Comparison: The Exact Bridge

Let Z μ fr denote a finite-type framed Higgs-incidence chart from Section 5. A comparison with S μ requires a geometric map to the minuscule Schubert chart or a coefficient correspondence implementing the same operation. In a chosen frame one may obtain a relative-position map
h : Z μ fr ⟶ Gr ¯ μ .
On the specified framed lattice chart the relative-position map is algebraic, as in Lemma 20. An extension to the genuine iterated Higgs target or to an unframed quotient still requires the corresponding geometric and equivariant data. A formal-disc trivialization alone does not give such a global unframed morphism.
Lemma 20
(The framed two-step lattice map exists). On the framed two-steplatticeincidence, the ordered lattice chain defines an algebraic morphism
h μ , μ : Z μ , μ Higgs ⟶ Gr ˜ μ , μ , ( A 2 , Λ 1 , Λ 2 ) ⟼ ( Λ 1 , Λ 2 ) .
Its composite with the terminal-lattice map is m ∘ h μ , μ . The map uses the chosen frame and does not by itself descend to unframed Higgs moduli. Pulling back a spherical coefficient along this map produces a well-defined MHMobject, but does not identify it with the kernel K 02 defined by (142).
Proof. 
The incidence scheme is a closed subscheme of B 2 × Gr ˜ μ , μ ; its second projection is therefore algebraic. The terminal map is a morphism of the projective lattice-chain moduli. Pullback on derived MHM exists for both morphisms and is functorial. To identify their pullbacks with the Higgs coefficient kernels, one must construct the relevant coefficient isomorphisms, not merely the underlying maps. □
Definition 33
(Restricted spherical kernel comparison). A restricted comparison consists of a defined algebraic map h as above, a specified choice of h * or h ! and its normalization, and an isomorphism
β μ : h † S μ → ∼ K μ
in D H ( Z μ fr ) , where h † denotes the chosen normalized pullback and K μ is the independently specified framed Higgs coefficient kernel of [4]. For two steps one further requires an isomorphism between the pullback/descent of S μ ★ S μ and the composed Higgs kernel, compatible with β μ on the two factors.
Remark 55
(Comparison does not follow from a common fiber). An isomorphism between individual exceptional P 1 fibers, or an identification of their top cohomology groups, does not produce β μ . Fedorov’s equivariant descent applies only after its group-action and finite-level hypotheses have been verified for the relevant comparison; an unframed quotient-stack operation is not supplied by the finite-type MHM theory alone. Only the selected kernel and its square are required for the following statements; no action of the entire spherical category is assumed.

6.7. Projecting the Actual Attachment to Both Channels

Assume a restricted two-step spherical kernel comparison has been constructed and carries the two idempotents of (163) to endomorphisms of the composed Higgs kernel. Assume these endomorphisms split in the selected idempotent- complete derived coefficient category. Let P sym and P det denote the corresponding natural summand projections of H 02 . Under these hypotheses define
ϵ sym : = P sym H 02 ( ϵ − ) , ϵ det : = P det H 02 ( ϵ − ) .
These morphisms have the correspondingly projected source and target objects; they are not elements of the original Ext MHM ( X 0 ) 1 group.
Proposition 53
(Channelwise decomposition of an attachment). Suppose the two kernel idempotents split the composed functor as H 02 ≃ H sym ⊕ H det , compatibly on objects and morphisms. Then
H 02 ( ϵ − ) ≃ ϵ sym ⊕ ϵ det .
In particular, H 02 ( ϵ − ) = 0 if and only if both projected morphisms vanish. Nonvanishing of the total arrow implies nonvanishing of at least one channel, but not of both.
Proof. 
Let e sym and e det be the two orthogonal idempotents, with sum the identity. By the assumed natural splitting, for every object M the value H 02 ( M ) is the direct sum of their images. Naturality gives
e j ( I − H ) [ 1 ] H 02 ( ϵ − ) = H 02 ( ϵ − ) e j ( R − H ) ( j = sym , det ) .
Consequently the off-diagonal components vanish, and the morphism is block diagonal with the two entries in (165). A direct sum of two morphisms is zero exactly when each component is zero. This proves the assertions. □

6.8. A Checkable Nonvanishing Statement Versus a Formal Decomposition

Definition 34
(Local detection of a projected arrow). Let U ⊆ X 2 be an admissible target open and ι : U ↪ X 2 its inclusion. For k ∈ { sym , det } , the projected morphism isdetected on Uif
ι * ϵ k ≠ 0 in D b MHM ( U ) .
When a stalk or costalk functor at a point of U gives a nonzero morphism, it supplies a sufficient detection test.
Proposition 54
(A sufficient local test). Under the hypotheses of Proposition 53, let y : { pt } ↪ U be a point. If y * ι * ϵ sym ≠ 0 , then ϵ sym ≠ 0 . The same conclusion holds with y ! in place of y * and with the determinant channel in place of the symmetric channel. The converse is not asserted for a selected point.
Proof. 
Both ι * and y * are functors on the relevant derived MHM categories and therefore send a zero morphism to zero. Thus a nonzero image under their composite forces the original morphism to be nonzero. The argument for y ! and the second channel is identical. A selected point need not detect every morphism in a derived category, so no converse follows. □
Remark 56
(Required local calculation). Application of Proposition 54 requires the actual target (co)stalk morphism, its perverse degree, and its Hodge and Tate normalization. The localized one-step nonvanishing of [4] does not compute these two-step channel maps. A zero determinant-channel calculation would be compatible with Proposition 53 and would not invalidate the two-step composition theorem.

6.9. Does Specialization Commute with Convolution? (Optional Compatibility)

Let G 0 , G 2 denote the chosen derived Rees specialization operations from Section 4 on the source and target coefficient categories. Suppose coherent-category Hecke functors H 01 , 0 and H 12 , 0 have been independently defined, with the appropriate intermediate specialization functor. Only after these data are fixed does it make sense to specify a natural transformation
σ M : G 2 H 12 H 01 M ⟶ H 12 , 0 H 01 , 0 ( G 0 M ) .
The map σ is not supplied by ordinary MHM base change, because its target is a different coefficient theory.
Definition 35
(Specialization–convolution defect). If (166) has been constructed as a natural transformation between exact functors to one stable coherent derived category, define
D σ ( M ) : = cofib ( σ M ) .
Its vanishing is equivalent to invertibility of σ M .
Proposition 55
(Compatibility on a distinguished triangle). Under the preceding hypotheses, applying σ to the source attachment triangle gives a morphism of distinguished triangles; the three cofibers of its components form a distinguished triangle in the target stable category.
Proof. 
The source and target of σ are exact functors, so each sends the attaching triangle to a cofiber sequence. Naturality of σ gives a morphism between these two cofiber sequences. Taking its objectwise cofiber yields a cofiber sequence in a stable category, because finite colimits commute and a square is cocartesian precisely when it is cartesian. Its three terms are the cofibers in Definition 35. □
Remark 57
(Independent status). Neither the definition of D σ nor Proposition 55 constructs σ. No comparison defect exists before the two exact functors and their transformation are defined. The local Rees specialization of Section 4 can be studied without this additional interchange problem.

6.10. The A 2 Collision as an Independent Stress Test

Recall the stabilized family from Section 2.8,
F s ( x , y , z , w ) = x 3 3 − s x + y 2 + z 2 + w 2 ,
whose critical-value curve is
T 2 = 4 9 s 3 .
The critical-point cover is étale away from s = 0 and ramified there; analytic continuation interchanges its two labeled branches. The category of its irregular Stokes data is not the category of mixed Hodge modules on the conifold, and an eigenprojector on its generic critical-value local system is not a Hecke kernel endomorphism.
Remark 58
(A separate comparison obligation). A comparison of (167) with (166) would require a specified geometric base map, compatible filtered Hecke kernels on the A 2 family, and a natural transformation between the resulting exact functors. Monodromy exchange of labels alone proves neither nonvanishing of D σ nor failure of the two-step MHM composition of Proposition 46.

7. Categorical Carriers for Hodge Atoms and the F-bundle Comparison Problem

The marked singular conifold Hodge atom of Definition 15 is an extension-bearing datum in mixed Hodge modules, defined through recollement and its specified attaching arrow rather than spectral decomposition. The KKPY spectral representatives below are separately defined objects, and no identification of the two notions is claimed.
The non-splitting theorem of Section 4 establishes survival of the conifold attaching morphism under associated-graded Rees specialization, and Theorem 1 identifies its global rational relation kernel with exceptional-curve relations when a projective small resolution exists. Its source is a singular regular-holonomic Hodge module, not a filtered flat vector bundle on the ambient fourfold. By contrast, the elementary Griffiths-transverse filtered-flat Rees construction below has zero leading Euler operator, although its base-direction Higgs field can be nonzero. We keep the two kinds of Rees construction separate and specify precisely the operations that a comparison with quantum spectral atoms would have to supply. The exceptional-curve identification is an equality of rational Hodge-theoretic maps; it does not itself construct a map from the singular filtered differential module to a quantum F-bundle or identify the corrected marked scalar.

7.1. Atomic Representatives, Spectral Markings, and Exactness

Fix a complete non-archimedean field k of characteristic zero, a smooth k -analytic base germ B F , and a parameter u at the origin of the analytic disk D u . The category Fbun k ( B F ) was fixed in Definition 7. Its objects are analytic bundles H on B F × D u with integrable meromorphic connection ∇ for which
∇ u ξ ( H ) ⊆ H , ∇ u 2 ∂ u ( H ) ⊆ H ( ξ ∈ T B F ) .
Morphisms are even horizontal morphisms on the specified base. KKPY show that this is a k -linear, additive, Karoubi-closed, quasi-abelian category [1]. Its standard exact structure consists of strict short exact sequences, not of arbitrary three-term sequences of horizontal morphisms.
Write
H 0 = H / u H , μ ξ = ∇ u ξ H 0 , κ = ∇ u 2 ∂ u H 0 .
These are well-defined operators on H 0 : if s is changed by u t , then the Leibniz rule gives ∇ u ξ ( u t ) = u ∇ u ξ ( t ) , since ξ ( u ) = 0 , while ∇ u 2 ∂ u ( u t ) = u 2 t + u ∇ u 2 ∂ u ( t ) . Thus both operators preserve u H . The operator κ is the leading Euler spectral operator; its sign relative to quantum multiplication is fixed by the convention for ∇ ∂ u .
Recall from Theorem 4.1 of [1] that the spectral splitting requires a maximal F-bundle at the chosen rigid point and a decomposition into invariant subspaces with pairwise disjoint spectra. In general the result is an external direct sum after a local splitting of the base. A horizontal idempotent on a common base arises only after applying the specified pullbacks and choosing the corresponding actual decomposition. An idempotent of the single fiber H ( b , 0 ) is not by itself such a morphism.
The marked category of Definition 8 is therefore used for selected spectral splittings. Its morphisms satisfy h e = e ′ h . The following calculation records the behavior of an extension when compatible idempotents have genuinely been constructed.
Proposition 56
(Projecting a strict F-bundle extension). Let
0 ⟶ A → a P → b C ⟶ 0
be a strict short exact sequence in Fbun k ( B F ) . Suppose horizontal idempotents e A , e P , e C are specified and satisfy
e P a = a e A , b e P = e C b .
Then the sequence splits as a direct sum of two strict short exact sequences, indexed by the idempotents e and 1 − e . If ϵ P : C → A [ 1 ] is its connecting morphism in the bounded derived category of the exact F-bundle category, then
e A [ 1 ] ∘ ϵ P = ϵ P ∘ e C .
In particular, the projected connecting arrow is defined on theactualspectral summands; the existence of e P alone, without (171), does not define such a projection.
Proof. 
Since Fbun k ( B F ) is Karoubi closed, the three idempotents split. Write P ≃ P e ⊕ P 1 − e , and similarly for A , C . The two identities in (171) say that a and b are block diagonal. Taking the image of e in the exact sequence (170) gives, on underlying bundles and at all strict filtration or lattice levels used by the exact structure,
0 ⟶ A e → a e P e → b e C e ⟶ 0 .
Exactness follows because the original strict sequence is the direct sum of its two block sequences; kernels and cokernels respect this direct-sum decomposition. Its other summand is obtained by replacing e by 1 − e . In the bounded derived category the original sequence provides a distinguished triangle A → P → C → ϵ P A [ 1 ] . The commuting pairs in (171) give an endomorphism of this triangle; naturality of the connecting morphism is precisely (172). □
Remark 59
(Atomicity is additional). A spectral summand becomes a KKPY G-atomic F-bundle only after the maximality, group-equivariance, parity and single-Euler-eigenvalue conditions in [1] have been verified. Its further equivalence class under [1] is a Hodge-atom invariant, not a new exact category. The marked singular conifold Hodge atom of Definition 15 has a persistent intersection-complex constituent, a ( − 2 ) -twisted punctual constituent, and a fixed connecting arrow. Its rigid/flexible terminology describes their respective roles without splitting the extension. This singular atom isnot identified with a KKPY spectral atom here; that would require a separately constructed extension-sensitive comparison as in Definition 37.

7.2. The Exact Source Category of Filtered Flat Bundles

The required input for a Rees-to-F-bundle construction is not an arbitrary regular-holonomic D Y -module. It is a vector bundle with an integrable connection and a locally split, Griffiths-transverse filtration on the base of that connection. We make this category explicit. Let k be a field of characteristic zero and B a smooth k-scheme. The same definitions can be made on a smooth formal or analytic base, when the corresponding locally free objects and connections exist.
Definition 36
(Griffiths-transverse filtered flat bundles). Let FilFlat k GT ( B ) have objects
( E , ∇ B , F • E ) ,
where E is a finite-rank vector bundle on B, ∇ B : E → E ⊗ Ω B / k 1 is an integrable connection, and F • E is a finite exhaustive separated decreasing filtration by subbundles, with locally free graded quotients, satisfying
∇ B ( F p E ) ⊆ F p − 1 E ⊗ Ω B / k 1 .
A morphism h : ( E , ∇ , F ) → ( E ′ , ∇ ′ , F ′ ) is a bundle map satisfying
h ( F p E ) ⊆ F ′ p E ′ , ( h ⊗ id ) ∇ B = ∇ B ′ h .
We use the strict exact structure: a sequence 0 → E A → E P → E C → 0 is a conflation when it is an exact sequence of flat bundles and, for every p, the induced sequence
0 ⟶ F p E A ⟶ F p E P ⟶ F p E C ⟶ 0
is exact with locally free quotients. No claim that the whole filtered category is abelian is made.
The identity map and composite of two morphisms satisfy (176), so this is an additive category. The standard pullback and pushout constructions for locally split strict sequences, with their induced connections and filtrations, give the stated exact structure. In particular, a composite of admissible monomorphisms remains admissible: applying the snake lemma to the filtration-level short exact sequences gives exactness of the composite filtration quotient, and local freeness is preserved by extensions of vector bundles. The dual statement applies to admissible epimorphisms. We will use only conflations of the explicitly described type.

7.3. A Constructed Rees-to-Formal-F-Bundle Functor

We now construct the promised functor, including its meromorphic connection and its action on morphisms. Let B ^ u = B × Spf k [ [ u ] ] , and let E ( ( u ) ) = E ⊗ O B O B ( ( u ) ) . Extend ∇ B to the punctured formal disk by the product connection
∇ ˜ = ∇ B + d u .
For the filtered bundle (174), define
R u ( E , F ) : = ∑ p ∈ Z u − p F p E [ [ u ] ] ⊆ E ( ( u ) ) .
Here F p E [ [ u ] ] denotes its u-adic scalar extension; the sum is a module over O B [ [ u ] ] . Because the filtration has finitely many jumps, it is locally finitely generated.
Lemma 21
(Local freeness and the special fiber). The lattice R u ( E , F ) is locally free of rank rk E over O B [ [ u ] ] . There is a canonical identification of O B -modules
R u ( E , F ) / u R u ( E , F ) ≃ ⨁ p ∈ Z gr F p E , gr F p E = F p E / F p + 1 E .
Proof. 
Work where all filtration subbundles and quotients have a common adapted local basis { e a } . Write p a for the largest index with e a ∈ F p a E . Thus F p E is spanned by those e a with p a ≥ p . Set g a = u − p a e a . For p ≤ p a one has u − p e a = u p a − p g a , and p a − p ≥ 0 . Consequently
R u ( E , F ) = ⨁ a O B [ [ u ] ] g a .
This proves local freeness. Reduction modulo u sends the class of g a to the class of e a in F p a E / F p a + 1 E . A change of adapted basis induces the ordinary associated-graded change of basis on the special fiber; therefore the identifications glue and are canonical. □
Theorem 6
(Exact functor to formal F-bundles). Let Fbun k for ( B ) denote the category of formal F-bundles on B ^ u satisfying the formal version of (168), as in [1]. The assignments
( E , ∇ B , F • E ) ⟼ ( R u ( E , F ) , ∇ ˜ ) ,
h ⟼ R u ( h ) , R u ( h ) ( u − p s ) = u − p h ( s ) ,
define a k-linear exact functor
R u : FilFlat k GT ( B ) ⟶ Fbun k for ( B ) .
In particular, it preserves the strict short exact sequences of Definition 36. The functor is faithful on morphisms. Its existence does not identify its target with the non-archimedean analytic category Fbun k ( B F ) .
Proof. 
Preservation by the connection. For s ∈ F p E and a vector field ξ on B, Griffiths transversality gives
∇ ˜ u ξ ( u − p s ) = u − ( p − 1 ) ∇ B , ξ ( s ) ∈ u − ( p − 1 ) F p − 1 E ⊆ R u ( E , F ) .
In the auxiliary direction s is independent of u, and hence
∇ ˜ u 2 ∂ u ( u − p s ) = − p u − p + 1 s = − p u ( u − p s ) ∈ u R u ( E , F ) .
The two operators also preserve arbitrary u-adic linear combinations of generators: their Leibniz terms are respectively u ξ ( f ) and u 2 ∂ u ( f ) , both regular for f ∈ O B [ [ u ] ] . These are exactly the formal F-bundle pole conditions. The product connection (178) is flat because ∇ B is integrable and has no u-dependence; in local coordinates [ ∇ ˜ ξ , ∇ ˜ η ] = ∇ ˜ [ ξ , η ] and [ ∇ ˜ ∂ u , ∇ ˜ ξ ] = 0 .
Morphisms. For h satisfying (176), h ( s ) ∈ F ′ p E ′ makes () an O B [ [ u ] ] -linear map of the Rees lattices. Horizontality in the base directions follows on generators from
∇ ˜ ξ ′ ( u − p h ( s ) ) = u − p ∇ B , ξ ′ h ( s ) = u − p h ( ∇ B , ξ s ) = R u ( h ) ∇ ˜ ξ ( u − p s ) .
It also holds in the u-direction, because h has no u-dependence. On generators R u ( id ) = id and R u ( k h ) = R u ( k ) R u ( h ) ; thus the functor axioms hold.
Strict exactness. Let 0 → E A → a E P → b E C → 0 be a strict conflation. By definition, for every p,
0 ⟶ F p E A → F p a F p E P → F p b F p E C ⟶ 0
is exact. If a sum of Rees generators maps to zero in R u ( E C , F C ) , localization at u and the exactness of the underlying sequence identify it with an element of E A ( ( u ) ) . Its membership in the Rees lattice of E A follows from strictness level by level in (188); equivalently one can work in a common adapted local basis and check every u-adic coefficient. Surjectivity follows by lifting each coefficient through the surjections F p E P → F p E C . We therefore obtain a short exact sequence of formal bundles with connection,
0 ⟶ R u ( E A , F A ) → R u ( a ) R u ( E P , F P ) → R u ( b ) R u ( E C , F C ) ⟶ 0 .
Locally this is a split short exact sequence of O B [ [ u ] ] -vector bundles, so its images and coimages agree in the strict exact structure of formal F-bundles. This proves exactness of R u .
Faithfulness. If R u ( h ) = 0 , invert u; the induced map on E ( ( u ) ) is h ⊗ 1 and therefore is zero. Since E and E ′ are locally free, this implies h = 0 . □
Corollary 7
(The connecting morphism of a Rees extension). For a strict short exact sequence in FilFlat k GT ( B ) , let ϵ F : E C → E A [ 1 ] denote its Yoneda connecting arrow. The exact functor R u induces an exact functor of bounded derived categories of the stated exact structures, and
ϵ TE : = D b R u ( ϵ F ) : R u ( E C , F C ) ⟶ R u ( E A , F A ) [ 1 ]
is the connecting arrow of (189). It is not asserted that ϵ TE ≠ 0 whenever ϵ F ≠ 0 .
Proof. 
For either exact category, a strict short exact sequence represents an element of Ext 1 and the associated distinguished triangle. The explicit three-term complex [ E A → a E P → b E C ] is acyclic for the strict exact structure. Applying R u termwise yields the acyclic complex of (189) by Theorem 6. Thus the induced derived functor maps the original triangle to the Rees triangle and maps the last arrow to (190). Equivalently, in the long exact sequence induced by Hom ( E C , − ) , the connecting homomorphism sends id E C to ϵ F . Naturality of the long exact sequences under the exact Rees functor sends this boundary to ϵ TE . There is no general assertion of injectivity on the resulting degree-one Hom group. □

7.4. The Euler Operator and an Explicit Filtered-Connection Model

The following calculation is decisive for comparison with the KKPY quantum F-bundle. On R u ( E , F ) define μ ξ and κ by (169).
Proposition 57
(Base-direction Higgs field and vanishing Euler operator). For the Rees F-bundle of Theorem 6, the special-fiber identifications (180) give
μ ξ | gr F p E : gr F p E ⟶ gr F p − 1 E , [ s ] ⟼ [ ∇ B , ξ s ] ,
whereas
κ R u ( E , F ) = 0 .
Thus the base-direction Higgs field can be nonzero even though the Euler spectral operator is identically zero.
Proof. 
If s ∈ F p E , equation (185) shows that the reduction of ∇ ˜ u ξ ( u − p s ) is represented by ∇ B , ξ s in F p − 1 E / F p E . Replacing s by s + t with t ∈ F p + 1 E changes ∇ B , ξ s by an element of F p E , so the map is independent of the representative. This proves (191). Equation (186) shows ∇ ˜ u 2 ∂ u ( R u ( E , F ) ) ⊆ u R u ( E , F ) , which proves (192) after reducing modulo u. □
For clarity, the same result can be read off from connection matrices. In an adapted basis e a of filtration index p a , write ∇ B , ξ e a = ∑ b Γ b a ( ξ ) e b . Griffiths transversality implies Γ b a = 0 unless p b ≥ p a − 1 . In the Rees basis g a = u − p a e a ,
∇ ˜ ξ g a = ∑ b Γ b a ( ξ ) u p b − p a g b , ∇ ˜ ∂ u g a = − p a u g a .
The only possible negative power in the base-direction matrix is u − 1 ; the auxiliary-direction matrix has at most a logarithmic pole. The coefficient of u − 1 in the former is exactly (191); the coefficient of u − 2 in the latter is zero.
Example 3
(A nonzero Higgs operator with zero Euler spectrum). Take B = Spec k [ t ] and E = O B e 1 ⊕ O B e 2 . Define
∇ B , ∂ t e 1 = e 2 , ∇ B , ∂ t e 2 = 0 , F 1 E = O B e 1 , F 0 E = E , F 2 E = 0 .
The connection is flat and ∇ B ( F 1 E ) ⊆ F 0 E ⊗ Ω B 1 . The Rees lattice has basis g 1 = u − 1 e 1 , g 2 = e 2 ; direct calculation gives
∇ ˜ ∂ t g 1 = u − 1 g 2 , ∇ ˜ ∂ t g 2 = 0 , ∇ ˜ u 2 ∂ u g 1 = − u g 1 , ∇ ˜ u 2 ∂ u g 2 = 0 .
Therefore, in the ordered basis ( [ g 1 ] , [ g 2 ] ) of the fiber at u = 0 ,
μ ∂ t = 0 0 1 0 ≠ 0 , κ = 0 0 0 0 .
This is an explicit formal F-bundle, not a quantum A -model F-bundle. Its base has dimension one while its fiber has dimension two, so it is not maximal in the sense of [1].
Proposition 58
(A spectral obstruction to a regular F-bundle isomorphism). Let F , Q be formal or non-archimedean analytic F-bundles on the same base, in a setting where both operators κ F and κ Q are defined. For every horizontal morphism h : F → Q one has
h 0 κ F = κ Q h 0 , h 0 = h | u = 0 .
If h is an isomorphism and κ F = 0 , then κ Q = 0 . In particular, anunshiftedRees bundle constructed by Theorem 6 cannot be isomorphic, through an F-bundle isomorphism regular and invertible at u = 0 , to a quantum F-bundle whose Euler operator is nonzero at the comparison point.
Proof. 
Horizontality gives h ∇ u 2 ∂ u F = ∇ u 2 ∂ u Q h . Both sides preserve the u-adic lattices. Reducing this equality modulo u yields (197). If h is invertible as an F-bundle morphism, then h 0 is an isomorphism; consequently κ Q = h 0 κ F h 0 − 1 = 0 . □
Lemma 22
(Exponential shifts and the scalar obstruction). Let L c be the rank-one F-bundle on the chosen base whose auxiliary-direction connection is ∇ ∂ u L c = ∂ u + c u − 2 and whose base-direction connection is trivial, for c ∈ k (or c ∈ k in the analytic setting). For every F-bundle F ,
κ F ⊗ L c = κ F + c id F 0 .
In particular, a single exponential shift of an elementary Rees F-bundle hasscalarEuler operator. It cannot be isomorphic to an F-bundle with a nonscalar Euler operator, even though its scalar eigenvalue need not vanish.
Proof. 
The tensor connection satisfies ∇ u 2 ∂ u F ⊗ L c = ∇ u 2 ∂ u F ⊗ 1 + 1 ⊗ ( u 2 ∂ u + c ) . Reduce modulo u to obtain (198). By Proposition 57, the Euler operator of the original Rees bundle is zero, hence after shifting it is c id . Intertwining under an isomorphism follows from Proposition 58. A scalar operator cannot be conjugate to a nonscalar one. □
Example 4
(A different, nonzero Euler spectral operator). Over the point base, the rank-one formal meromorphic connection
∇ ∂ u = ∂ u − λ u − 2 , λ ∈ k × ,
on k [ [ u ] ] satisfies the formal F-bundle condition because ∇ u 2 ∂ u = u 2 ∂ u − λ is regular. Its Euler operator on the special fiber is κ = − λ ≠ 0 . Proposition 58 therefore excludes an F-bundle isomorphism from an elementary Rees bundle to this object. The rank-one example does not purport to be a quantum F-bundle of a specified smooth variety.
Remark 60
(The actual quantum connection). In the conventions of [1], the formal quantum connection contains ∇ ∂ u = ∂ u − u − 2 ( Eu ★ − ) + u − 1 Gr , so that its special-fiber operator is κ Q = − ( Eu ★ − ) . This expression also contains base-direction quantum multiplication, Gromov–Witten data, and in general a different analytic parameter space. If its Euler multiplication is nonzero, the obstruction in Proposition 58 applies. KKPY also allow exponential shifts by rank-one F-bundles; Lemma 22 shows that a shift changes κ by a scalar. Thus the unshifted obstruction is not invariant under all such shifts, while a nonscalar Euler operator remains incompatible with anysingleshifted elementary Rees bundle. If the resulting Euler operators agree, an isomorphism of connections, equivariant structures and spectral markings still has to be constructed. A zero Euler eigenvalue is not a proof of atomicity or of extension preservation.
Remark 61
(The Calabi–Yau small-quantum Euler locus). In the convention of Remark 60, the quantum Euler operator is κ Q = − ( Eu ★ − ) . For a smooth projective Calabi–Yau variety Y with c 1 ( T Y ) = 0 , consider its ordinary, non-equivariant, graded small-quantum base, with parameters only in H 2 ( Y ) . The coefficient of every degree-two coordinate in the standard conformal Euler vector field is 1 − deg ( H 2 ) / 2 = 0 , and the constant contribution c 1 ( T Y ) also vanishes; coordinates of other degrees are zero on this locus. Consequently Eu and κ Q vanishthroughoutthe small-quantum locus, not only at its origin:
Eu H 2 ( Y ) = 0 , κ Q H 2 ( Y ) = 0 .
The statement uses precisely this non-equivariant grading and small-parameter restriction; equivariant parameters, bulk variables of different degrees, or another Euler convention require an independent check. Vanishing eliminates the particular mismatch between κ R u = 0 and a nonzero quantum Euler operator. It does not identify the full meromorphic connections, base-direction operators, coefficient fields, parameter bases, spectral markings, or extension morphisms. Neither the exceptional-curve relation theorem nor the Lee–Lin–Wang matrix comparison constructs the missing functor (210).
Proposition 59
(Exceptional-curve relations and the conifold-transition matrix). Suppose X 0 is a proper nodal complex threefold admitting a projective small resolution ρ : X ˜ 0 → X 0 with smooth source and exceptional curves C 1 , … , C k . Let V Σ H = ⨁ p ∈ Σ Q H ( − 2 ) and let G ρ : V Σ H → H 4 ( X ˜ 0 , Q ) be the curve-class map of (89). Under the fixed marking of Section 3,
L Σ , corr H = D λ − 1 ker G ρ .
Write μ = dim Q ker G ρ and choose a rational matrix A ∈ Mat k × μ ( Q ) whose columns form a basis of ker G ρ . Then G ρ A = 0 and im A = ker G ρ ; in particular the matrix of corrected relations, in the fixed node basis, is D λ − 1 A . The transpose A t : Q k → Q μ instead has kernel ( ker G ρ ) ⊥ , with respect to the standard node-coordinate pairing. It is therefore adifferent mapfrom G ρ or the canonical connecting morphism. For a projective conifold transition satisfying the hypotheses of Lee–Lin–Wang [13], their exact sequence of weight-two Hodge structures and logarithmic-connection comparison use this relation-matrix formalism after complexification and their independently fixed choices of vanishing-cycle and curve-class coordinates. The identification of relation spaces (201) does not identify those connections, nor does it imply an extension-preserving F-bundle comparison.
Proof. 
The first equality is precisely the signed boundary–Gysin comparison of Theorem 1; its minus sign has no effect on the kernel, while the diagonal marking D λ remains. Choosing a basis of ker G ρ gives G ρ A = 0 and im A = ker G ρ by construction. Since D λ is invertible over Q , its inverse maps these basis columns to a basis of the corrected kernel. The standard linear-algebra identity ker A t = ( im A ) ⊥ proves the transpose assertion. The comparison with the Lee–Lin–Wang exact sequence uses only its curve-relation matrix and does not identify its quotient by im B with H 4 ( X ˜ 0 ) : its domain, target, and Hodge-weight normalization are independently specified by the transition. □

7.5. Two Existing Constructions That Do Not Yet Compose

We now compare the precise types of the constructions available from the singular central fiber and from the smooth family. Let ι : X 0 ↪ Y be the fixed smooth ambient embedding of Section 2.3, and let R Y = R F ord D Y be its Rees algebra. Saito’s filtered-component functor and the algebraic Rees functor give the actual composite
MHM ( X 0 ) → For F , ι FilRH str ( Y ) → R z GrMod ( R Y ) ,
with the exact structures and derived refinements specified in Section 2 and Section 4; the right-hand category here denotes graded R Y -modules, with the Rees images lying in its coherent graded subcategory. The auxiliary variable here is z. These modules are defined on the smooth ambient space Y and carry its differential-operator action; when M has singular support they are not generally vector bundles with nonsingular flat connection on an independently specified base. There is no arrow from the right-hand category of (202) to FilFlat k GT ( B ) provided merely by Saito strictness, Rees specialization, or the identification of the characteristic support in T * Y .
Proposition 60
(The local conifold Rees model is not an ambient flat bundle). Let Y = A C 4 and X det = { x w − y v = 0 } , and let P ˜ loc denote the filtered localization module identified with the canonical conifold extension in Section 4. Its underlying left D Y -module is nonzero and supported on the proper closed hypersurface X det . It is therefore not the D Y -module of a nonzero locally free bundle carrying an integrable nonsingular connection on Y. Consequently its Hodge Rees module cannot be inserted as an object of FilFlat C GT ( Y ) , or of the domain of Theorem 6, by identifying the filtration parameters z and u.
Proof. 
The local identification gives the underlying localization quotient O Y [ g − 1 ] / O Y with its nonzero stalks along X det and zero restriction to Y ∖ X det . A nonzero vector bundle of constant positive rank on the connected scheme Y has support all of Y; the same is true for its associated D Y -module with nonsingular integrable connection. Therefore the two underlying D Y -modules cannot be isomorphic. A change of Rees parameter cannot change the underlying support or manufacture a smooth-base flat connection. This proves the asserted type obstruction; it does not exclude a different, independently constructed nearby-cycle or Gauss–Manin realization. □
Remark 62
(Integrable Rees modules are a different admissible target). The support obstruction in Proposition 60 excludesnonsingular flat vector bundles on Y; it does not exclude a category of integrable Rees or R Y -modules permitting holonomic singular support. The canonical localization quotient already has such an algebraic Rees module, and its extension survives specialization by Theorem 3. The grading operator on a Rees module can be used to define a candidate auxiliary parameter-direction action, but it does not alone establish a KKPY F-bundle: a prescribed connection must satisfy the u-Leibniz rule, the pole and integrability conditions, and compatibility with a specified base and coefficient comparison. No categorical equivalence or identification of z with u is claimed here.
There is, however, a separate geometric source of filtered flat bundles in a smooth proper family. The relation-matrix comparison of Proposition 59 is not a bridge between these source categories. Let
π × : X × ⟶ B ×
be the restriction of the degeneration to a smooth base B × = B ∖ { 0 } , under the additional hypothesis that π × is smooth and projective. For each integer m the sheaf V Q m = R m π * × Q is a polarizable variation of pure Hodge structure. Let
E m = V Q m ⊗ Q O B × , ∇ GM : E m → E m ⊗ Ω B × 1 , F • E m
be its flat bundle with Gauss–Manin connection and Hodge filtration. The filtration is locally by subbundles, and Griffiths transversality holds:
∇ GM ( F p E m ) ⊆ F p − 1 E m ⊗ Ω B × 1 .
Consequently, with coefficient field k = C , and after choosing a compatible formal model of the smooth complex base (or the corresponding complex-analytic formal variant), Theorem 6 constructs the formal F-bundle
T m ( π × ) : = R u ( E m , ∇ GM , F • E m ) over B × × Spf k [ [ u ] ] .
This is a construction from the smooth family; it is not an object assignment on arbitrary mixed Hodge modules on X 0 .
Proposition 61
(The smooth-locus restriction loses the attachment). Let j : U = X 0 ∖ Σ ↪ X 0 and let ϵ Σ : R Σ H → I H [ 1 ] be the corrected arrow. Then
j * R Σ H = 0 , j * ϵ Σ = 0 .
For any exact functor T : D b MHM ( U ) → E into a stable category, the composite T ∘ j * sends ϵ Σ to zero. Therefore an exact realization that detects ϵ Σ cannot factor through j * alone.
Proof. 
The object R Σ H is the finite direct sum ⨁ p ∈ Σ i p * Q H ( − 2 ) , supported on the complement of U. Recollement gives j * i p * = 0 . Hence j * R Σ H = 0 . Apply the exact functor j * to R Σ H → ϵ Σ I H [ 1 ] ; its source is the zero object and its image is the zero morphism. An exact functor T sends this zero morphism to zero. The factorization conclusion follows. □
Remark 63
(Nearby smooth fibers are a different operation). Proposition 61 does not say that the Gauss–Manin variation in (204) forgets every singularity: its monodromy and limiting Hodge data can depend on the degeneration. It says only that a functorfactoring through the restriction of a central-fiber MHM to Uannihilates the specified punctual arrow. The nearby- and vanishing-cycle maps
can H : Ψ π H → Φ π H , var H : Φ π H → Ψ π H ( − 1 )
are defined for appropriately chosen mixed Hodge modules on the total family, with the normalization of Section 2.5. Their relation to the fixed extension ϵ Σ is determined by the corrected variation-cone gluing; converting these maps into an F-bundle morphism requires a further defined operation. Neither (202) nor (206) supplies it automatically.

7.6. The Exact Comparison Required for an Atomic Realization

We formulate the smallest statement that would turn a categorical proximity into a realization. Fix a geometric subcategory C π of D b MHM ( X 0 ) on which the corrected extension is defined and which is specified independently of its proposed image. For example, any construction using the total family must supply its restriction/specialization maps and a rule for morphisms, not merely selected cohomology vector spaces. Let B F be a fixed non-archimedean analytic base or a germ, and let
D F ( B F ) : = D ex b Fbun k ( B F )
be the bounded derived category formed using the canonical strict exact structure. This target already has a source of morphisms, shifts, distinguished triangles and extension classes.
Definition 37
(Extension-sensitive atomic comparison datum). Anatomic comparison datumfor the corrected conifold extension consists of:
(i)
a geometrically specified exact functor
R at : C π ⟶ D F ( B F ) ,
with a construction on objects and morphisms and verified preservation of identities, composition, and triangles;
(ii)
an independently defined strict extension or distinguished triangle in the target category,
A I → a Q A P → b Q A R → ϵ Q A I [ 1 ] ;
(iii)
endpoint isomorphisms η I : R at ( I H ) → ∼ A I and η R : R at ( R Σ H ) → ∼ A R that are obtained independently of the desired equality of arrows;
(iv)
a specified comparison of parameter bases and coefficients, with actual spectral projectors and equivariant structures whenever the chosen A ’s are required to be KKPY atomic factors.
It isextension-sensitiveif the endpoint comparisons satisfy
η I [ 1 ] ∘ R at ( ϵ Σ ) = ϵ Q ∘ η R ,
with a coherent enhancement of the square whenever a higher-functorial statement is asserted.
The phrase “independently defined” matters: an arbitrary target morphism ϵ Q could always be chosen by transporting R at ( ϵ Σ ) through arbitrarily chosen isomorphisms. That would not identify the conifold class with the extension intrinsic to a smooth geometric construction. The purpose of (212) is to test two previously specified constructions.
Proposition 62
(Boundary and comparison square). Assume the data of Definition 37 are constructed, and suppose there is a morphism of distinguished triangles from the image of I H → P H → R Σ H → ϵ Σ I H [ 1 ] to (211), with endpoint components η I , η R . Then (212) holds. If both endpoint components are isomorphisms, the middle component is an isomorphism. This statement does not construct the functor or the morphism of triangles.
Proof. 
A morphism of distinguished triangles commutes with each of the four structure maps, including the connecting map. The fourth square is exactly (212). To show the middle component is an isomorphism, apply Hom ( T , − ) for any target object T to the two triangles. We obtain a morphism of long exact sequences. For every degree, the four terms adjacent to the Hom group of the middle objects map isomorphically by the endpoint hypotheses. The five lemma therefore makes the middle Hom map an isomorphism in each degree. The Yoneda criterion in the triangulated homotopy category gives the claimed isomorphism of middle objects. □
Remark 64
(The scalar and the spectral identification are different tests). The canonical conifold comparison in Section 3 produces a specific scalar or quotient endomorphism relating twoMHMextensions. The Rees specialization class of Section 4 belongs to a graded-module derived category. The arrow ϵ Q of (211) belongs to the derived category of F-bundles. These are different Hom groups. Equality of the local MHM scalars, survival under z = 0 , and identification of a quantum atomic factor are three distinct questions. A diagram linking them requires the actual functors and natural transformations; it is not inferred from equality of dimensions, spectral covers, or exceptional fibers.

7.7. A Precise Interface for the Subsequent Bivariant Theory

The elementary construction above gives a genuine and reusable result: the exact Rees-to-formal-F-bundle functor (184), together with the explicitly calculated operators (191) and (192). The non-splitting at z = 0 proved in Section 4 is a separate result in the category of associated-graded differential modules; neither it nor the formal F-bundle construction identifies the two categories. A comparison with a specified quantum F-bundle requires the additional operations below. In particular, the global curve-relation identification (201) is a genuine result on the singular side, but does not supply a morphism into an analytic quantum bundle.
First, one must construct, on a geometric domain including the corrected extension, the missing operation that turns the relevant singular or relative MHM data into a Griffiths-transverse filtered flat bundle or directly into an analytic F-bundle. The nearby- cycle maps, their monodromy, and the normalization of the corrected variation-cone arrow must be included in the object and morphism assignment if the punctual extension is to remain detectable. Second, the resulting formal bundle must be brought to the same coefficient field and parameter base as the independently defined KKPY quantum bundle; if an analytic comparison is used, its convergence and base-change hypotheses are part of the construction. Third, the comparison must intertwine the full connections, not only the fiber dimensions or the base-direction Higgs fields. At u = 0 this entails the necessary operator identities
h 0 μ ξ src = μ ξ Q h 0 , h 0 κ src = κ Q h 0 ,
for vector fields ξ identified by the actual base map. Fourth, the marked attaching square (212) must be calculated on the independently defined endpoints, and a claim of detection requires R at ( ϵ Σ ) ≠ 0 . Finally, if the object is called an atomic factor, the maximality, G-equivariance, parity, and eigenvalue hypotheses of [1] must be checked, with any spectral projectors compatible with the transported extension as in Proposition 56.
Remark 65
(A single KKPY atom does not determine an MHM splitting). KKPY show that their chemical formula has a single atom in the nef-canonical setting [1]. A smooth Calabi–Yau fiber can therefore carry a single KKPY atom even if the degeneration has several distinct Hodge-theoretic subspaces or a nonsplit punctual MHM extension. Neither their number nor their extension classes can be identified solely by counting KKPY spectral factors. A comparison respecting (212), rather than a numerical identification, is needed.
Remark 66
(Boundary of the established results). The category Fbun k ( B F ) and its spectral decomposition under KKPY’s hypotheses are imported theorems. Proposition 56, Theorem 6, and Propositions 57, 58, 61, and 59 are proved in this section with their stated restrictions. A functor of type (210), an analytic comparison to the specified quantum connection, and the equality (212) are not established by those results. They constitute the exact extension-sensitive comparison problem to be tested on the chosen degeneration, without altering the internal MHM–Hecke calculations of Section 5 and Section 6.

8. Higgs and Magnetic Hecke Tests

The preceding calculations permit a local Higgs modification to be tested without supposing that a gauge-theoretic correspondence has already been lifted to a derived coefficient category. This section applies the invariant-line and first-jet calculations of Section 5 and Section 6 to the magnetic Hecke interface, retaining the scheme structure of semisimple, nonsemisimple, and scalar fibers and the effect of successive modifications. The underlying two-step incidence is calculated in Theorem 4; the arguments below identify which parts of that geometry survive in rational and coherent coefficients and which additional maps a magnetic comparison requires. We then specify the mathematical content of the extended-Bogomolny scattering correspondence and exactly what it does and does not imply for an extension-bearing Hodge-module object.
The physical motivation involves two different theories. Conifold vanishing cycles occur in the N = 2 type-IIB compactification setting; magnetic Hecke modifications and Wilson/’t Hooft operators occur in the Kapustin–Witten N = 4 gauge-theory setting. We do not identify their Hilbert spaces, supersymmetry algebras or categories of line operators. The precise interface considered here is the independently constructed rank-two Higgs Hecke correspondence and its action on the already specified MHM attaching morphism. Scattering geometry supplies another description of the compatible modifications under its own analytic hypotheses; a categorical identification needs additional maps.

8.1. The Invariant-Line Calculation

Let V be a two-dimensional complex vector space with a chosen basis, and let A ∈ End ( V ) represent the leading value of a Higgs field at the marked point of C, after choosing a local trivialization of K C there. Its invariant-line scheme is
Inv ( A ) = { ℓ ∈ P ( V ) : A ℓ ⊂ ℓ } ,
with the closed subscheme structure defined by vanishing of the induced homomorphism L → ( V / L ) , where L = O P ( V ) ( − 1 ) is the tautological line. Write
φ = A = a b c d .
For homogeneous coordinates [ s : t ] on P ( V ) , the induced homomorphism is represented by the determinant of the column vectors ( s , t ) T and A ( s , t ) T . Thus
q A ( s , t ) : = c s 2 + ( d − a ) s t − b t 2 = 0 .
Both the equation and its scheme structure are invariant under a change of basis, up to the natural action on the projective coordinates.
Proposition 63
(Complete classification of invariant-line schemes). Put
Δ ( A ) = ( a − d ) 2 + 4 b c = ( tr A ) 2 − 4 det A .
Then the scheme Inv ( A ) has exactly the following forms over C :
(i)
If Δ ( A ) ≠ 0 , it consists of two distinct reduced points, the two eigenspaces of A.
(ii)
If Δ ( A ) = 0 but A is not scalar, it is a single nonreduced scheme of length two, supported on the unique eigenline.
(iii)
If A = λ id V , the equation is identically zero and Inv ( A ) = P 1 scheme-theoretically.
In the first two cases Inv ( A ) is a divisor of degree two on P 1 ; in the scalar case its dimension jumps to one.
Proof. 
An eigenline ℓ = C ( s , t ) is equivalent to linear dependence of the two columns in the definition; this yields (215) as an equality of closed subschemes. The discriminant of the binary quadratic q A equals ( d − a ) 2 + 4 b c , which is also the characteristic-polynomial discriminant. When Δ ( A ) ≠ 0 , the binary quadratic has two distinct projective zeros; since its derivative does not vanish at either, the corresponding divisor is reduced. When Δ ( A ) = 0 and A is not scalar, q A is a nonzero square of a linear form, up to a scalar, so it defines a length-two double point. Finally, q A ≡ 0 forces c = b = d − a = 0 , hence A = a I , and conversely such a matrix fixes every line. This exhausts the cases. □
Example 5
(Semisimple, nilpotent, and scalar charts). For A s s = diag ( λ 1 , λ 2 ) with λ 1 ≠ λ 2 , the equation is ( λ 2 − λ 1 ) s t = 0 , so the scheme is the disjoint union of [ 1 : 0 ] and [ 0 : 1 ] . On the affine chart s ≠ 0 , u = t / s is a local equation u = 0 near [ 1 : 0 ] , with no infinitesimal direction.
For the regular nilpotent matrix A n = 0 1 0 0 , the equation is − t 2 = 0 . In the same chart,
Inv ( A n ) ∩ { s ≠ 0 } = Spec C [ u ] / ( u 2 ) .
Its support is [ 1 : 0 ] , but its local algebra has dimension two and a nonzero square-zero tangent parameter u. A nonsemisimple matrix λ I + A n has the same invariant-line scheme. The zero and nonzero scalar matrices instead give the whole P 1 .
Proposition 64
(What the two coefficient theories detect). Let L = Inv ( A ) for the three cases above. For a constant rational Betti coefficient on the underlying analytic space, the only nonzero cohomology groups of L are:
L H 0 ( L , Q ) H 2 ( L , Q ) { p 1 , p 2 } Q ⊕ 2 0 Spec C [ u ] / ( u 2 ) Q 0 P 1 Q Q ( − 1 ) .
In contrast, the coherent structure sheaf of the doubled point has H 0 ( L , O L ) = C [ u ] / ( u 2 ) , of complex dimension two; for two reduced points H 0 ( L , O L ) ≃ C 2 , and for P 1 it is C . Consequently, the constant Betti (or its rational mixed-Hodge) realization cannot by itself recover the nilpotent element of the doubled-point structure sheaf.
Proof. 
The analytic topological space of a finite Artinian scheme is the finite discrete set of its closed points. Therefore a pair of points has H 0 = Q 2 and no positive cohomology, while the doubled point has the topology of a single point and gives H 0 = Q . The usual cellular cohomology of P 1 ( C ) ≃ S 2 is H 0 = Q and H 2 = Q ( − 1 ) , the latter notation recording its pure Hodge structure. The coherent computations are the global-section rings of the finite affine schemes and H 0 ( P 1 , O ) = C . In the nonreduced case the square-zero element u survives in the structure sheaf but is invisible in the underlying topological space. □

8.2. First Jets, Incidence Degeneration, and Hitchin Invariance

The scheme-theoretic difference in the preceding proposition is also present in the two-step Higgs incidence of Section 6. The next calculations record the corresponding local chart and the unchanged Hitchin polynomial; the full punctual incidence and its cohomology are already established in Theorem 4. Let O = C [ [ w ] ] and consider
Λ 1 = w O e 1 ⊕ O e 2 , A − ( w ) = a ( w ) w β ( w ) c ( w ) d ( w ) .
The first lattice is A − -invariant by the divisibility of its upper-right entry. In the adapted basis ( w e 1 , e 2 ) , the induced Higgs matrix is
A + ( w ) = a ( w ) β ( w ) w c ( w ) d ( w ) .
The two matrices are conjugate over C ( ( w ) ) by the change of lattice basis D = diag ( w , 1 ) :
A + = D − 1 A − D .
Here D is not invertible over O , so this is a conjugacy over the punctured formal disc, not an automorphism of the original holomorphic lattice.
Proposition 65
(First-jet fiber classification). Let a 0 = a ( 0 ) , d 0 = d ( 0 ) , and β 0 = β ( 0 ) ; equivalently β 0 is the first coefficient of the original upper-right entry. The scheme of admissible second lines in Λ 1 / w Λ 1 ≃ C 2 is
L 2 = V v ( ( d 0 − a 0 ) u − β 0 v ) ⊂ P [ u : v ] 1 .
It has two distinct reduced points when d 0 ≠ a 0 , a length-two doubled point when d 0 = a 0 and β 0 ≠ 0 , and the complete P 1 when d 0 = a 0 and β 0 = 0 . In particular, two original Higgs matrices with the same residue can give different second-step incidence schemes.
Proof. 
The determinant for the two vectors ( u , v ) and A + ( 0 ) ( u , v ) , using (219), is
det u a 0 u + β 0 v v d 0 v = ( d 0 − a 0 ) u v − β 0 v 2 .
This gives (221). If d 0 − a 0 ≠ 0 , the factors v and ( d 0 − a 0 ) u − β 0 v have different projective zeros. If d 0 = a 0 and β 0 ≠ 0 , the equation is − β 0 v 2 , a doubled point. If both coefficients vanish, the equation vanishes identically. For instance, the matrices A − ( 0 ) = 0 and A − ( 1 ) = 0 w 0 0 have the same residue but β 0 = 0 and 1, respectively; these yield P 1 and the doubled point. □
Proposition 66
(Whole reduced first-jet incidence over a punctual source). Assume A ( 0 ) = 0 and write A ( w ) = w B + O ( w 2 ) . Let F ≃ F 2 be the full two-step lower-lattice surface, with first-line projection π : F → P 1 and distinguished section S contracted by the terminal-lattice map. The Higgs-invariant two-step incidence over this fixed first jet is the closed scheme cut out, in adapted coordinates, by
− β B ( ℓ ) v 2 = 0 , β B ∈ H 0 ( P 1 , O ( 2 ) ) ,
where β B ( ℓ ) is the induced map from the first line to its quotient. Its reduction is (i) S together with two vertical fibers when B has two distinct eigenlines; (ii) S together with one vertical fiber when B is nonscalar Jordan; and (iii) the entire F 2 when B is scalar. Its respective rational Hodge cohomology is
( H 0 , H 2 , H 4 ) = ( Q , Q ( − 1 ) 3 , 0 ) , ( Q , Q ( − 1 ) 2 , 0 ) , ( Q , Q ( − 1 ) 2 , Q ( − 2 ) ) .
This whole-fiber classification is distinct from the second-line fiber over a fixed first lattice in Proposition 65. For nonscalar B, the full scheme is generically doubled along S by the factor v 2 ; the displayed cohomology is calculated only after reduction. For scalar B the full incidence is the reduced surface F 2 , not a doubled section. These are precisely the cases of Theorem 4.
Proof. 
The first-order matrix B induces a homomorphism ℓ → C 2 / ℓ , a section of O ( 2 ) . After the first lower modification and reduction modulo w, the transformed Higgs matrix is strictly upper triangular with upper-right entry β B ( ℓ ) . Its invariant-line equation is − β B ( ℓ ) v 2 = 0 . For a non-eigenline the reduction permits only the distinguished second line v = 0 , which forms S. At every eigenline, the whole vertical second-line fiber appears. The zero divisor of a nonscalar matrix’s eigenline polynomial has two distinct points or a repeated point; for scalar B it vanishes identically. In the first two cases the reduced rational curves form trees, so Mayer–Vietoris gives one degree-two Tate class per component and no odd cohomology. In the scalar case the cellular cohomology of F 2 gives the displayed groups. □
Remark 67
(Correction of a potential scalar-case ambiguity). The full-fiber condition is d 0 = a 0 and β 0 = 0 ; there is no condition that a 0 itself vanish. The third case in Proposition 65 means only equality of the diagonal coefficients and vanishing of β 0 , not a restriction to the zero matrix. A scalar nonzero leading matrix also fixes every second line.
Proposition 67
(A nonflat family visible from the incidence equation). For A − ( τ ) ( w ) = 0 τ w 0 0 , let L ⊂ A τ 1 × P [ u : v ] 1 be the second-line incidence family. Its equation is τ v 2 = 0 . The generic fiber is the doubled point V ( v 2 ) , while the fiber at τ = 0 is P 1 . The morphism L → A τ 1 is not flat; on u ≠ 0 its ring is C [ τ , z ] / ( τ z 2 ) and z 2 ≠ 0 is annihilated by τ.
Proof. 
Apply Proposition 65 with β 0 = τ , a 0 = d 0 = 0 , giving τ v 2 = 0 . On u ≠ 0 put z = v / u ; then τ z 2 = 0 . The element z 2 is nonzero in the quotient but is killed by τ , so the algebra has C [ τ ] -torsion. A module over the principal ideal domain C [ τ ] is flat only if it is torsion-free; thus flatness fails. The fibers are obtained by setting τ to a nonzero value or zero in the equation. □
Proposition 68
(Preservation of the Hitchin characteristic polynomial). For each admissible lower elementary Higgs modification with A + as in (219), one has
tr ( A + ) = tr ( A − ) = a + d , det ( A + ) = det ( A − ) = a d − w β c .
Thus the characteristic polynomial is unchanged. If these local matrices arise from global holomorphic Higgs bundles identified away from x, their Hitchin invariants agree globally. The same holds for two consecutive admissible modifications. This geometric invariance does not imply equality of the sheaf-theoretic kernels or of their images of ϵ − .
Proof. 
Compute the diagonal sum and determinant from both displayed matrices. Alternatively, conjugacy (220) over the punctured disc gives equality of characteristic polynomials there; their coefficients are regular, so they are equal over O . Globally the characteristic coefficients are holomorphic sections of the appropriate tensor powers of K C . The modification identifies them on the dense open C ∖ { x } ; a holomorphic section vanishing on that open vanishes everywhere. For two modifications, apply the argument successively. No coefficient-functor identity was used. □

8.3. Extended Bogomolny Scattering and Its Actual Hypotheses

Let W = [ 0 , 1 ] × C with a specified product-type metric and an interior singular point ( y 0 , x ) , where x ∈ C is the Hecke point. Write W ∘ = W ∖ { ( y 0 , x ) } . Let ( E , H ) be a Hermitian rank-two bundle on W ∘ , let A be a unitary connection, ϕ ∈ Ω 1 ( W ∘ , u ( E ) ) , and ξ ∈ Ω 0 ( W ∘ , u ( E ) ) . In the convention of [14], the extended Bogomolny equations take the form
F A − 1 2 [ ϕ ∧ ϕ ] = * d A ξ , d A ϕ − * [ ξ , ϕ ] = 0 , d A * ϕ = 0 .
The singularity has a prescribed Dirac cocharacter type k = ( k 1 , k 2 ) , with its sign convention fixed relative to the lower-modification coweight μ . Near the puncture the solution is compared, in a chosen admissible singular gauge, to the Dirac model of type k ; boundary values at { 0 , 1 } × C and the gauge group at those boundaries are fixed as part of the moduli problem. The paper [14] establishes a Kobayashi–Hitchin correspondence between the appropriately constrained solutions of this PDE and Hecke modifications of Higgs bundles; the analytic metric, asymptotic and stability conditions in that result must be retained when invoking it.
Definition 38
(Scattering correspondence for fixed analytic data). Let M μ EB ( W ∘ ) denotethe moduli setof gauge-equivalence classes of solutions to (224) satisfying the specified Dirac, boundary and regularity conditions in the cited Kobayashi–Hitchin setting. Let H μ Higgs be the corresponding set of isomorphism classes of admissible Hecke modifications ( E − , φ − ; E + , φ + ; Λ + ⊂ Λ − ) with the matching coweight convention. Thescattering map S μ associates to a solution the pair of end Higgs bundles and the meromorphic identification induced by parallel transport in the interval direction, with its local Hecke lattice at x.
Proposition 69
(The geometric content of scattering). For fixed data satisfying the hypotheses of [14], scattering identifies the stated Kobayashi–Hitchin moduli problem with its corresponding admissible Higgs Hecke-modification problem. In the rank-two lower minuscule case, any corresponding lattice has a line ℓ ⊂ ( E − ) x satisfying φ − ( x ) ℓ ⊂ ℓ , hence it lies in the scheme (215) at the level of closed points. This geometric identification alone does not induce an equivalence of the MHM coefficient categories, an isomorphism of their derived extensions, or an identification of the nonreduced scheme structure in (217) with the analytic deformation theory of the PDE.
Proof. 
The first sentence invokes the Kobayashi–Hitchin correspondence in [14] for its stated Dirac and boundary hypotheses; we do not extend it to arbitrary boundary conditions. For the second sentence, the lower minuscule lattice is the inverse image of a line in Λ − / w Λ − by Proposition 31. Holomorphic extension of the modified Higgs field is equivalent to preservation of that lattice, hence to φ − ( x ) ℓ ⊂ ℓ by Lemma 8. Therefore the line is a closed point of (215). The scattering theorem is stated for its own gauge-theoretic moduli problem; it does not construct an MHM functor or identify the scheme structure of the invariant-line fiber with a PDE-derived deformation complex. The last assertions cannot be inferred from the preceding geometric calculation. □
Remark 68
(Gauge-theoretic deformations are separate). The Zariski tangent line of the doubled point Spec C [ u ] / ( u 2 ) is a first-order deformation of aHiggs-invariant line. It is not, without a comparison of deformation complexes, the cokernel of a linearized extended-Bogomolny operator. Likewise, a bijection on isomorphism classes cannot be used to transport a Yoneda class: it contains no morphism-space assignment and no compatibility with cones.

8.4. The Common Mathematical Carrier of the Two Physical Routes

The conifold light-sector description from type-IIB compactification starts with three-cycles in a smoothing and their associated charges. The marked singular conifold Hodge atom of Definition 15 retains a specific Hodge-theoretic attaching arrow but supplies no physical identification with magnetic line-operator sectors. The Kapustin–Witten description of magnetic line operations starts with cocharacters and Hecke modifications of bundles or Higgs bundles. They are physically different inputs; nothing in the calculation above equates a wrapped D3-brane state with a Wilson or magnetic line operator. What is presently common to our mathematical analysis is the selected Hecke geometry and its functor on D H ( S − ) .
More precisely, let Arr ( D H ( S + ) ) = Fun ( Δ 1 , D H ( S + ) ) be the existing arrow category. The corrected attachment on the framed source is the actual arrow ϵ − : R − H → I − H [ 1 ] , and the Hecke correspondence constructed earlier produces
H μ ( ϵ − ) : H μ ( R − H ) ⟶ H μ ( I − H ) [ 1 ] .
The two-step operation carries this arrow to the corresponding composed morphism by Corollary 6. This is a mathematical carrier of an attachment within MHM; it is not yet a physical realization functor. The punctual first-jet calculation of Theorem 4 determines the reduced incidence and constant-coefficient endpoints, whereas strict punctual stalk restriction of the conifold attachment is zero by Proposition 48. The corrected class is instead detected locally by its nonzero costalk coordinate. Lemma 19 provides a costalk fiber formula for the genuine composed correspondence, and Theorem 5 identifies the additional comparison and nonvanishing hypotheses. Neither the reduced punctual table nor the costalk formula alone evaluates the numerical boundary on the full twice-modified Higgs target. The formal filtered-flat F-bundle functor of Theorem 6 belongs to a separate coefficient category; scattering geometry does not identify it with the transported mixed-Hodge-module arrow or construct a magnetic realization functor.
Remark 69
(Smooth-locus restriction and the punctual arrow). Suppose an alleged magnetic or spectral realization of the conifold attachment factors through restriction to U = X 0 ∖ Σ . It sends ϵ Σ to zero. Thus such a factorization cannot be anextension-detectingcomparison of the original singular arrow with a nonzero magnetic or atomic target arrow.
Remark 70
(No inference from the whole first-jet fiber). The reduced punctual F 2 incidence and its terminal-lattice constant-coefficient direct images do not encode the actual coefficient morphism on the genuine twice-modified Higgs target. The extraordinary fiber formula of Lemma 19 provides the correctly oriented detection test, but its reduction to the nonzero ODP costalk coordinate requires the verified comparison and kernel exchange of Theorem 5. In particular, neither the two Satake tensor-square channels nor the numerical perverse boundary maps are determined by the constant-coefficient cohomology table. A stronger magnetic statement would also require a correspondence-level comparison compatible with scattering and the specified attaching map.

8.5. A Conditional Magnetic Action Comparison

To make a magnetic comparison meaningful, we must distinguish two possible levels of structure. At the geometric level, scattering relates admissible solutions of (224) to Higgs modifications. At the coefficient level, an attachment-sensitive comparison would require functors on objects and morphisms and a natural transformation between their Hecke operations.
Definition 39
(A typed magnetic comparison). Assume an independently constructed stable coefficient category D mag on the selected gauge-theoretic moduli problem, an exact magnetic correspondence operator T μ on it, and an exact functor R mag into the chosen MHM or Betti target category E . Suppose that the Higgs Hecke operator H μ acts on E and the domains agree. A magnetic comparison is aconstructednatural transformation
Θ μ mag : R mag T μ ⟹ H μ R mag .
The geometric scattering map does not itself define the functor R mag or the transformation Θ μ mag .
Proposition 70
(The precise attachment-level magnetic test). If the above comparison has been constructed and α : C → A [ 1 ] is an arrow in its domain, the square
R mag T μ ( C ) → R mag T μ ( α ) R mag T μ ( A ) [ 1 ] Θ μ , C mag ↓ Θ μ , C mag Θ μ , A mag [ 1 ] ↓ Θ μ , A mag [ 1 ] H μ R mag ( C ) → H μ R mag ( α ) H μ R mag ( A ) [ 1 ]
commutes. To identify its lower arrow with (225) requires explicit endpoint isomorphisms and compatibility with thespecifiedconifold class, not just a gauge-theoretic matching of Hecke-modification parameters.
Proof. 
Apply naturality of Θ μ mag to α . Because each functor is exact, its component on A [ 1 ] is the suspension of its component on A, giving the displayed square. The arrow (225) has already been defined from ϵ − in the MHM category. Identifying two arrows in that category requires the endpoint isomorphisms to make the marked square commute, precisely as in Definition 13. This is not implied by naturality alone. □
Remark 71
(What an analytic scattering equivalence would give). If, in addition to the pointwise Kobayashi–Hitchin correspondence, an actual isomorphism of the selectedanalytic moduli spaces(or suitable analytic stacks) is established, it induces inverse pullback/pushforward equivalences on the corresponding Betti sheaf categories. To compare theirHeckefunctors one must still verify the Cartesian squares of the scattering map with the two correspondence projections and identify the coefficient kernels. No such derived or stack-level isomorphism is asserted merely from the published bijection of the stipulated gauge moduli and Higgs modifications.
Remark 72
(Electric–magnetic duality remains a stronger problem). To compare a magnetic operator with a Wilson operator would further require an independently constructed duality functor and a natural transformation intertwining the two, including the two-step fusion constraints. Neither the determinant incidence calculation nor the extended-Bogomolny scattering map establishes the N = 4 electric–magnetic duality equivalence of [15]. The restricted result here is the fully calculated Higgs invariant-line geometry, its Hitchin preservation, and the explicit type of the missing coefficient comparison, which are testable inputs for a later line-operator realization.
Remark 73
(Scope of the Section 8 results). The rank-two invariant-line scheme, its finite-jet refinement and the Hitchin-polynomial calculation are algebraic statements. The extended-Bogomolny correspondence is imported with its analytic Dirac, boundary and regularity hypotheses. A functorial identification of magnetic and MHM/Betti attaching arrows remains conditional on a category of coefficients, a geometric comparison of correspondences, a kernel morphism, and compatibility on the marked arrow. These conditions are carried to the proof-status ledger of Section 9 without being asserted as a result of scattering alone.

9. Synthesis, Limitations, and the Bivariant Continuation

The marked singular conifold Hodge atom A Σ H = ( I H , R Σ H , ϵ Σ ; u , v ) of Definition 15 records the persistent intersection-complex constituent, the punctual constituent, and their specified, potentially nonsplit attachment in the existing mixed-Hodge-module category. Its canonical recollement version and the corrected marked version must be distinguished by their fixed quotient normalizations. The conifold localization model of Theorem 2 and the non-splitting result of Theorem 3 establish survival of the nonzero corrected local extension under associated-graded Rees specialization at z = 0 , without fixing its marked variation-cone scalar. The canonical boundary also admits the Verdier-dual description of Proposition 27. For projective nodal threefolds with a smooth projective small resolution, Theorem 1 identifies the global attaching boundary, including its rotation sign, with the exceptional-curve Gysin map. In a distinct explicit calculation, Theorem 4 determines the full punctual two-step first-jet incidence, its reduced Hodge cohomology, and its terminal-lattice images. These are results about specified mixed-Hodge-module objects, connecting morphisms, and geometric maps, rather than numerical substitutes for them. Open–closed recollement fixes the canonical boundary; framed Hecke operations transport the selected arrow; and a verified Beck–Chevalley square identifies two-step composition.
A second, independently developed construction concerns the categorical carriers of Hodge atoms in the sense of Katzarkov–Kontsevich–Pantev–Yu (KKPY). Theorem 6 constructs an exact functor on Griffiths-transverse filtered flat bundles, with actual morphisms and strict extension sequences. It also gives an operator-level obstruction to identifying the elementary Rees output with an independently specified quantum F-bundle. There is not yet a functor from the singular mixed-Hodge-module attachment to that quantum F-bundle category. Keeping the two constructions distinct makes their possible comparison mathematically testable without making it an unproved premise of the mixed-Hodge-module results.
We record the operations available to a subsequent bivariant construction, the global relation exact sequences and their exceptional-curve realization, and an R1–R8 proof-status ledger. The smooth-purity theorem of Section 6 detects a genuinely transported two-step arrow under concrete hypotheses; verification of those hypotheses on the actual Higgs charts and calculation of its numerical perverse boundary remain distinct problems. The final subsections formulate the extension-sensitive comparison and the restricted bivariant transport to which the proved calculations apply.

9.1. Bivariant Hodge Atoms

Let a : X 0 → Spec C be the proper structure morphism of the algebraic nodal central fiber, and let Σ be its finite node set. Retain the normalized exact sequence from Section 2:
I H = IC X 0 H , R Σ H = ⨁ p ∈ Σ i p * Q H ( − 2 ) , I H → u P H → v R Σ H → ϵ Σ I H [ 1 ] .
The first line specifies objects of MHM ( X 0 ) and the second is a distinguished triangle in D b MHM ( X 0 ) . Together they constitute the corrected marked singular conifold Hodge atom of Definition 15; the analogous canonical atom uses the recollement arrow ϵ can and the canonical endpoint maps. Its Yoneda class and endpoint maps are part of the chosen datum, imported from [2]; in particular, the quotient Q H ( − 2 ) is not the Q H ( − 1 ) contribution of the normalized vanishing-cycle interface. We do not replace the marked extension by an arbitrary generator of a local one-dimensional extension space.
Definition 40
(Admissible transport record). Anadmissible transport recordfor (228) consists of finite-type complex algebraic spaces X − , X 1 , X 2 and aspecifiedsource arrow ϵ − : R − H → I − H [ 1 ] in D H ( X − ) . Its provenance is either an algebraic map e − : X − → X 0 together with the equality ϵ − = e − * ( ϵ Σ ) and its endpoint identifications, or an algebraic determinant-chart map e 0 : X − → X det together with the independently normalized local attachment ϵ det and a specified comparison between that local input and the relevant restriction of (228). Similarity of analytic ordinary-double-point germs alone is not such a comparison. The record further contains two algebraic correspondences
X − ← p 01 Z 01 → q 01 X 1 , X 1 ← p 12 Z 12 → q 12 X 2 ,
kernels K 01 ∈ D H ( Z 01 ) and K 12 ∈ D H ( Z 12 ) , and their specific cohomological shifts and Tate twists. The chosen operations must preserve the bounded coefficient categories; the Cartesian square for the fiber product Z 01 × X 1 Z 12 must admit the indicated Beck–Chevalley and projection-formula isomorphisms. For a statement on an unframed quotient, the record also contains effective descent data, including equivariance of the selected attaching arrow.
A record isverifiedwhen the displayed schemes, maps, kernels, normalizations, and required exchange isomorphisms have been constructed or follow from cited results with hypotheses satisfied on the stated geometric charts. No extension to arbitrary nodal families or Higgs moduli spaces is part of the definition.
Proposition 71
(Transport and composition of the marked arrow). For a verified record define, with the exact star convention,
H i j ( M ) = q i j * K i j ⊗ L p i j * M , i j ∈ { 01 , 12 } .
Let Z 02 = Z 01 × X 1 Z 12 , with projections r 01 , r 12 , and let H 02 be the functor of the composed correspondence with kernel r 01 * K 01 ⊗ L r 12 * K 12 . Then each H i j sends the specified source triangle to a distinguished triangle and there is a natural equivalence
κ H : H 12 H 01 → ∼ H 02 .
On the connecting morphism it satisfies the equality
κ I − H H [ 1 ] ∘ H 12 H 01 ( ϵ − ) = H 02 ( ϵ − ) ∘ κ R − H H .
The equality belongs to the Hom group in the target derived category; it does not assert nonvanishing or concentration in the perverse heart.
Proof. 
The evaluation pullback carries the input triangle to a distinguished triangle on X − , with last arrow ϵ − . The functors p i j * , tensor with K i j and q i j * are exact functors of the chosen stable coefficient categories. The transported last arrow is therefore exactly H i j ( ϵ − ) .
For clarity, write the Cartesian square as
Z 02 → r 01 Z 01 r 12 ↓ r 12 q 01 ↓ q 01 Z 12 → p 12 X 1 .
Its specified Beck–Chevalley isomorphism is p 12 * q 01 * → ∼ r 12 * r 01 * . The projection formula for r 12 gives, naturally in M,
H 12 H 01 ( M ) = q 12 * K 12 ⊗ p 12 * q 01 * ( K 01 ⊗ p 01 * M ) → ∼ q 12 * K 12 ⊗ r 12 * r 01 * ( K 01 ⊗ p 01 * M ) → ∼ q 12 * r 12 * r 12 * K 12 ⊗ r 01 * K 01 ⊗ ( r 01 * p 01 * M ) → ∼ H 02 ( M ) .
This is the natural equivalence constructed, with its units, counits, and exchange transformations, in Proposition 46. Naturality at ϵ − : R − H → I − H [ 1 ] gives (231); exactness identifies the component at I − H [ 1 ] with the shift of the component at I − H . No step invokes a t-exactness or a Hom-faithfulness property. □
Remark 74
(Imported and newly computed geometric information). The kernel-line small-resolution component and the full invariant-line incidence are distinguished in Proposition 32; the latter is generically two-to-one, and the former has an exceptional P 1 . The two-step chain surface and first-jet invariant-line schemes are calculated in Propositions 40 and 41, and the entire reduced punctual incidence is computed in Theorem 4. These are actual local geometric results. They do not replace the additional verification that a chosen global Higgs-modification diagram is the selected fiber product with the stated kernels. Likewise, the localized survival and higher-degree results imported from [4] retain their specified target opens, normalizations, and distinction between star and shriek images. The pure projective G L 2 Schubert calculation in [3] carries a punctual Q H ( − 1 ) contribution on a different base; it is not the conifold quotient R Σ H .
Definition 41
(Verified specialization datum). Fix a closed embedding i : X 0 ↪ Y into a smooth complex algebraic variety. Aspecialization datumconsists of the strict filtered image of the exact sequence underlying (228), the induced Rees modules A ˜ , P ˜ , C ˜ and their specified connecting morphism ϵ R : C ˜ → A ˜ [ 1 ] in the graded Rees-derived category. Its specialized morphism is
ϵ 0 = ϵ R ⊗ C [ z ] L C [ z ] / ( z ) : sp 0 C ˜ ⟶ ( sp 0 A ˜ ) [ 1 ] .
The record includes a claim of vanishing or nonvanishing only when the actual morphism has been evaluated in the indicated target Hom group; strictness alone is not such an evaluation.
Remark 75
(Specialization is a separate calculation). Section 4 proves strictness, constructs the Rees connecting class, and identifies the kernel of the specialization map as the image of multiplication by z in a derived Hom group (Proposition 25). At a determinantal node, Proposition 29 further expresses the actual extension through four homogeneous cocycle components a i = x i m , with relations x i a j − x j a i = 0 and a simultaneous four-equation coboundary test after reduction modulo z. The canonical localization presentation and first Hodge levels of Theorem 2 and Lemma 6 now determine the conifold outcome: Theorem 3 proves the specialized extension is nonsplit. The four-component Koszul representative remains a chain-level test, not an outstanding prerequisite for the nonvanishing conclusion. The two finite filtered models of Section 4.5 illustrate why strictness alone is insufficient. No Rees–Hecke interchange theorem is assumed.
Theorem 7
(Recollement and functorial properties of marked singular conifold Hodge atoms). Let X 0 be the algebraic nodal threefold of Definition 15, with fixed coefficients, cone convention, and endpoint markings. Let A Σ , can H and A Σ H denote its canonical and corrected marked atoms. The following conclusions hold, with the hypotheses of the individual constructions retained.
(i)
Local filtered survival.On the determinantal ordinary-double-point chart, the canonical short exact sequence has a nonsplit associated-graded filtered- D -module sequence. Equivalently, its Rees connecting arrow is nonzero after derived specialization at z = 0 . The conclusion also holds for a corrected local atom whose specified attaching class is nonzero, without assuming that its marked scalar λ p equals 1.
(ii)
Global exceptional-curve boundary.Suppose X 0 is proper and admits a projective small resolution ρ : X ˜ 0 → X 0 with smooth source and exceptional curves C p ≃ P 1 . With the complex-oriented Gysin map G ρ ( e p ) = PD [ C p ] and the specified small-map identification θ ρ , one has
θ ρ ∂ Σ , can H = − G ρ , θ ρ ∂ Σ , corr H = − G ρ D λ , L Σ , corr H = D λ − 1 ker G ρ ,
provided the corrected local classes are nonzero, so that D λ is invertible over Q .
(iii)
Derived Hecke composition and conditional detection.For a verified transport record of Definition 40, the selected evaluation pullback of the atom is carried by each exact pull–tensor–push operator to a specified distinguished triangle, and the Beck–Chevalley/projection-formula equivalence identifies its two-step image with the image under the composed correspondence, compatibly with the marked connecting morphism. If, in addition, the actual two-step correspondence satisfies the properness, nonempty-fiber, smooth evaluation composite, node-factorization, and normalized constant-kernel hypotheses of Theorem 5, and its local costalk coordinate is nonzero, the extraordinary target costalk and hence the transported derived attaching morphism are nonzero.
The assertion in(iii)does not establish those additional hypotheses on every genuine Higgs chart and does not evaluate individual perverse boundary morphisms. No assertion identifies the marked singular atom with a KKPY spectral atom or constructs a functor to quantum F-bundles.
Proof. 
For (i), the canonical recollement extension of Proposition 9 is identified on the determinant chart with the filtered localization model of Theorem 2. Theorem 3 proves that its associated-graded sequence does not split, and Proposition 15 transports this non-splitting under any invertible corrected local quotient marking. This is a claim about the particular local model and not about arbitrary strict filtered extensions.
For (ii), the extraordinary costalk computation and purity counit in Theorem 1 identify the canonical local arrow with the negative of the complex-oriented exceptional-curve Gysin arrow. Proper direct image gives the first equality of (233); the diagonal-rescaling law of Proposition 18 gives the second, and invertibility of D λ gives the stated kernel. The conclusion is rational and does not determine an integral relation lattice or the individual marked scalars.
For (iii), the specified evaluation map first pulls the attaching triangle into D H ( X − ) ; it is not implicitly an arrow on a different base. Proposition 46 and Corollary 6 construct the composition equivalence and its naturality square on that exact arrow. Lemma 19 expresses its extraordinary target costalk through the genuine target fiber and coefficient morphism. Under the extra assumptions in Theorem 5, smooth purity, extraordinary-pullback composition, and dualizing cohomology of the nonempty proper fiber prove that costalk morphism is nonzero. No perverse-degree numerical conclusion is used. The three claims are therefore compatible properties of the designated atom under separately specified operations, not an assertion that their distinct coefficient categories have been identified. □
Remark 76
(The non-spectral meaning of the construction). The atom of Definition 15 is specified by a recollement-derived extension and its marked morphism, rather than by a decomposition into eigenspaces of an Euler or spectral operator. Its persistent and punctual constituents are not asserted to be a direct sum, and the word “atom” introduces no new source or target category. The separately constructed categorical representatives of KKPY spectral atoms in Section 7 are not identified with this singular object; an extension-sensitive comparison between them remains the independent construction problem (244).

9.2. The Eight Core Calculations and Their Precise Outputs

The principal output of the mixed-Hodge-module part is not a universal comparison theorem. It is a selected package of explicit morphisms, local calculations, and naturality statements. Their status is recorded here at the level necessary to use them without changing their hypotheses.
(R1)
Canonical gluing. Definition 15 fixes the marked singular atom, while the localization triangle constructs P can H = H 0 p ( j * j * I H ) and a canonical short exact sequence with quotient R Σ H (Proposition 9). The adjunction-induced map P H → P can H and its quotient endomorphism reduce equality with the marked variation-cone extension to a definite normalization check (Proposition 10). The marked equality is not inferred merely from isomorphism of the endpoint objects.
(R2)
Local normalization. The ODP costalk and derived adjunction give
Hom ( i p * Q H ( − 2 ) , I H [ 1 ] ) ≃ Q .
See Corollary 2. The canonical recollement boundary has costalk coordinate − 1 and is the negative of the complex-oriented exceptional-curve Gysin generator when a projective small resolution exists (Theorem 1). The separate marked scalar λ p comparing the corrected variation-cone class to that canonical boundary remains uncomputed; the Gysin sign does not force λ p = 1 .
(R3)
Filtered specialization. Saito strictness supplies the filtered and Rees connecting morphisms. The Bockstein kernel criterion is proved and the local four-generator Koszul resolution explicitly computes the complex in which the conifold class must be represented (Propositions 25 and 29). The conifold specialization is nevertheless decided: Theorem 3 proves non-splitting by the annihilator of the punctual degree-two generator in gr 2 F P can ≃ O X , o . For any nonzero corrected local class the same conclusion follows by invertible endpoint rescaling; the exact marked scalar remains unresolved. The four-component cocycle is retained as an independent explicit representative.
(R4)
One-step Hecke transport.Section 5 fixes the framed evaluation, the two incidence components, and the derived kernel operation. The transported arrow is the image of the designated source arrow, not a newly selected extension generator. Comparison with an independently normalized kernel and descent to an unframed target require additional hypotheses.
(R5)
Two-step Higgs incidence. The two-step lattice chain is F 2 , while Higgs admissibility cuts out a smaller, first-jet-dependent incidence scheme. Its second-line fibers have two distinct reduced points, a doubled point, or a whole P 1 ; properness and the corresponding constant-coefficient fiber cohomology are calculated. The entire punctual incidence is computed in Theorem 4; its nonreduced scheme structure is retained separately from the cohomology of its reduction, and the terminal-lattice images are determined in Proposition 43. Their interpretation on genuine Higgs-moduli target charts requires the relevant geometric comparison of incidence spaces.
(R6)
Two-step attaching arrow. The composition equivalence (230) and its action on ϵ − are established whenever the coefficient base-change hypotheses are verified (Proposition 46 and Corollary 6). The full numerical perverse boundary δ 02 n remains uncomputed. Proposition 48 explains why strict punctual stalk restriction loses the attachment, whereas Lemma 19 gives its genuine target-costalk formula. Theorem 5 proves that this derived costalk arrow is nonzero if the actual target fiber is nonempty and proper, the source evaluation composite is smooth near that fiber and factors through the node on it, and the composed kernel has the specified shifted constant form. Smooth purity and dualizing cohomology then replace an assumed nonzero pushforward by a proof. The verification of these geometric hypotheses on the actual Higgs charts, the individual maps δ 02 n , and any Satake-channel nonvanishing are separate tasks.
(R7)
Global node relations. The actual derived-global-sections boundary ∂ Σ H and its kernel are constructed in Section 3.9. The rational diagonal-rescaling law is proved in Proposition 18. For a projective small resolution with smooth source, the boundary–Gysin comparison is proved, with the canonical sign fixed: θ ρ ∂ Σ , can H = − G ρ and θ ρ ∂ Σ , corr H = − G ρ D λ . Thus L Σ , corr H = D λ − 1 ker G ρ (Theorem 1). No integral lattice equality or projective small resolution for every nodal threefold is implied. A morphism of derived global attachment triangles induces a map on these kernels (Proposition 19); the selected bivariant geometric map is additional data and is not manufactured by an arbitrary relation matrix.
(R8)
Restricted coherence. The Beck–Chevalley and projection-formula maps and their units and counits are written in Section 6. Its octahedral calculation includes chain maps, a connecting map and a null homotopy (Proposition 51). Internal MHM two-step naturality is established; coherence with an independently defined Satake, magnetic, or quantum atomic functor requires its actual kernel or comparison transformation.
Theorem 7 packages these results as distinct, explicitly restricted properties of the marked singular conifold Hodge atom; it does not create a new comparison functor. The list distinguishes proven identities, theorems whose explicit hypotheses still require checking for a proposed Higgs application, and numerical calculations not yet made. In particular, the local specialization nonvanishing of Theorem 3 does not imply nonvanishing after twice-Hecke transport: the latter is a different coefficient morphism on a different target, addressed by the sufficient smooth-purity theorem only under its stated hypotheses.
Proposition 72
(Global node relations as an exact quotient). Let V Σ H = ⨁ p ∈ Σ Q H ( − 2 ) , and let ∂ Σ H : V Σ H → I H 4 ( X 0 , Q ) be the connecting map defined by (80)– (81). Put L Σ H = ker ∂ Σ H . Then there are canonical isomorphisms in MHS Q
L Σ H ≃ coker H 0 ( a * I H ) ⟶ H 0 ( a * P H ) , coker ∂ Σ H ≃ H 1 ( a * P H ) .
In particular, the global extension gives the two short exact sequences
0 ⟶ L Σ H ⟶ V Σ H ⟶ im ∂ Σ H ⟶ 0 , 0 ⟶ im ∂ Σ H ⟶ I H 4 ( X 0 , Q ) ⟶ H 1 ( a * P H ) ⟶ 0 .
These statements are intrinsic to the chosen extension and do not require a small resolution or a separate tropical model.
Proof. 
Apply the exact derived functor a * to (228). Because a * R Σ H = V Σ H is concentrated in degree zero, the long exact sequence of mixed Hodge structures contains
H 0 ( a * I H ) ⟶ H 0 ( a * P H ) → H 0 ( a * v ) V Σ H → ∂ Σ H I H 4 ( X 0 , Q ) ⟶ H 1 ( a * P H ) ⟶ H 1 ( a * R Σ H ) = 0 .
Exactness at H 0 ( a * P H ) and V Σ H identifies the cokernel of the first map with the image of the second, which is ker ∂ Σ H . Exactness at I H 4 ( X 0 , Q ) and H 1 ( a * P H ) identifies the cokernel of ∂ Σ H with H 1 ( a * P H ) . The two short exact sequences are the kernel–image and image–cokernel sequences of the indicated morphism in the abelian category MHS Q . □
Corollary 8
(Exceptional-curve realization of the global exact sequences). Assume that X 0 admits a projective small resolution ρ : X ˜ 0 → X 0 with smooth source and exceptional curves C p , and let θ ρ and G ρ be normalized as in Theorem 1. For the corrected extension there are canonical isomorphisms of rational Hodge structures
L Σ , corr H = D λ − 1 ker G ρ , im ∂ Σ , corr H → θ ρ ∼ im G ρ , H 1 ( a * P corr H ) ≃ H 4 ( X ˜ 0 , Q ) / im G ρ .
Hence the two exact sequences of (235) identify, after the fixed small-map comparison, with the kernel–image and image–cokernel sequences of the exceptional-curve class map precomposed by the invertible marked diagonal D λ . The result is rational; it neither specifies an integral saturation nor identifies the corresponding quantum connection.
Proof. 
Theorem 1 identifies the actual global connecting morphism with − G ρ D λ . Its kernel is D λ − 1 ker G ρ , and its image becomes im G ρ because D λ is invertible. Proposition 72 identifies H 1 ( a * P corr H ) with the cokernel of that connecting map. Transporting the cokernel through θ ρ gives the last isomorphism and the exact sequences. The argument uses no integral identification. □
The exceptional-curve relation space admits a rational basis matrix A with G ρ A = 0 and im A = ker G ρ . The corresponding corrected relation columns are D λ − 1 A . As made precise in Proposition 59, the transpose of this relation matrix is a different morphism from G ρ ; this distinction is retained when comparing with the conifold-transition exact sequence of Lee–Lin–Wang [13]. Neither a comparison of quantum connections nor a canonical integral relation matrix follows from the rational boundary–Gysin theorem.
Corollary 9
(Naturality of the relation quotient). Suppose two specified nodal attachments are connected by an actual morphism of their derived global attachment triangles. Denote its maps on punctual and intersection-cohomology terms by u V : V Σ H → V Σ ′ H and u I : I H 4 ( X 0 , Q ) → I H 4 ( X 0 ′ , Q ) . Then
u I ∘ ∂ Σ H = ∂ Σ ′ H ∘ u V , u V ( L Σ H ) ⊆ L Σ ′ H ,
and these maps induce a morphism between both short exact sequences in (235).
Proof. 
A morphism of derived triangles commutes with their connecting maps. Applying H 0 after proper direct image gives the first identity. For ℓ ∈ L Σ H , ∂ Σ ′ H ( u V ℓ ) = u I ( ∂ Σ H ℓ ) = 0 , proving the inclusion. The maps of images and cokernels follow from commutativity of the square and exactness of the rows. □

9.3. Criteria for a Completed Result

Definition 42
(Verification data and proof status). For a mathematical claim T in this paper, its verification record is
V ( T ) = ( Geom , C , D , α , Φ , η , Hyp , Calc ) .
Here Geom identifies the actual varieties, stacks, strata and maps; C , D are its source and target categories; α is the specified input arrow; Φ is the constructed functor or correspondence operation; η is any additional comparison morphism or natural transformation; Hyp contains all geometric and categorical hypotheses, including shifts, Tate twists and exact structures; and Calc comprises the maps, exact sequences, chain-level identities and computations used in its proof.
A claim isproved on its stated domainwhen these data are specified and its hypotheses verified or explicitly included in the statement. It isconditional for a proposed applicationwhen a required geometric comparison or hypothesis has not been verified for the intended chart. It iscomputationally unresolvedwhen the relevant arrow exists but its actual image has not been evaluated. If no candidate operation has been constructed on objects and morphisms, a proposed comparison is aconstruction problem, not an established functor.
Proposition 73
(Compatibility, detection, and coverage). Let Φ : C → D be an exact functor of stable categories, and let α : C → A [ 1 ] be a specified arrow. Suppose β : C ′ → A ′ [ 1 ] is independently defined in D . If isomorphisms η C : Φ ( C ) → ∼ C ′ and η A : Φ ( A ) → ∼ A ′ satisfy
η A [ 1 ] ∘ Φ ( α ) = β ∘ η C ,
then the marked attachments are identified on the chosen object. Moreover Φ ( α ) ≠ 0 if and only if β ≠ 0 . The existence of Φ on a geometric class does not imply (239), and verification of that equality for one arrow does not extend the domain of Φ.
Proof. 
The first assertion is the stated equality of morphisms after identifying their domains and codomains. Inverting the endpoint isomorphisms gives Φ ( α ) = η A [ 1 ] − 1 β η C ; therefore one of these two arrows vanishes exactly when the other does. No conclusion follows for an arbitrary target arrow without the independently constructed endpoint maps and equality. The last statement is a distinction of quantifiers: checking a property of one element of a source Hom group constructs no new objects or morphisms in the domain of the functor. □
Remark 77
(Four comparison targets retain their original names). The original comparison targets F1–F4 of Section 1.6 remain useful as a separate interface to the R1–R8 calculations. F1 asks for an independent source category and a genuinely constructed comparison with the specified marked arrow; F2 asks for the numerical Rees specialization; F3 asks for an independently defined Hecke-kernel intertwiner; and F4 asks for compatibility of its two-step compositions. Here F2 is resolved for the actual nonzero local conifold class by Theorem 3; the calculation is not a general faithfulness theorem for Rees specialization. F3 and F4 have internal MHM analogues that do not construct the cross-category transformations. The sufficient smooth-purity nonvanishing criterion of Theorem 5 strengthens R6, but the genuine chart hypotheses and degreewise numerical boundaries remain to be checked. R7 is now identified with exceptional-curve relations under the projective small-resolution hypothesis, not merely under a postulated Gysin comparison. F1 does not follow from using the arrow category of D b MHM ( X 0 ) as a new name for the original data. The KKPY comparison adds a specifically typed, separate realization problem discussed below.

9.4. What the Restricted Calculation Provides

In the terminology of Definition 15, the restricted bivariant input is the marked singular atom and its images under verified operations, as summarized by Theorem 7. The first reusable output for the bivariant sequel is a morphism in the arrow category, together with the chosen canonical composition equivalence, not a numerical “extension character”. The second output is a degreewise criterion for testing its perverse placement once the target coefficient objects and the actual costalk coefficient morphism are known. The ordinary stalk of the strict punctual attachment is zero, while the genuine target costalk has a proved smooth-purity nonvanishing criterion. Consequently punctual constant-coefficient dimensions cannot replace the required check of that criterion on the actual Higgs target, or the subsequent degreewise perverse calculation.
Proposition 74
(A restricted bivariant transport test). Fix a verified two-step transport record and let F = H 02 and G = H 12 H 01 . The pair
F ( ϵ − ) , κ H | ϵ −
is well defined, with κ H | ϵ − the commutative square (231) in the arrow category. Thus the result of one composed correspondence and the result of two successive correspondences represent the same marked arrowunder the specified equivalence, not as literally equal morphisms in distinct source presentations.
If H k denotes standard cohomology for the MHM t-structure (equivalently, perverse cohomology after applying rat), then the transported triangle supplies the long exact sequence
… ⟶ H k F ( I − H ) ⟶ H k F ( P − H ) ⟶ H k F ( R − H ) → δ k H k + 1 F ( I − H ) ⟶ H k + 1 F ( P − H ) ⟶ … ,
where δ k = H k F ( ϵ − ) , with the target shift understood. If any δ k is nonzero, then F ( ϵ − ) ≠ 0 . The same implication holds for a nonzero image of F ( ϵ − ) under a chosen stalk, costalk, or other exact test functor. No converse is asserted for a single such test.
Proof. 
The first assertion follows by evaluating the natural equivalence κ H : G → ∼ F on ϵ − : R − H → I − H [ 1 ] . Its naturality square is the arrow-category morphism, and both vertical components are isomorphisms. Applying the cohomology functors of the standard t-structure to the distinguished triangle F ( I − H ) → F ( P − H ) → F ( R − H ) → F ( ϵ − ) F ( I − H ) [ 1 ] gives (241). Its connecting map is induced by the explicitly displayed last arrow, not chosen independently. A functor sends a zero morphism to zero. Hence a nonzero cohomological, stalk, or costalk image forces F ( ϵ − ) ≠ 0 . A morphism can be invisible to a particular cohomology or stalk test, so vanishing of that test does not imply vanishing of the derived arrow. □
Remark 78
(Negative calculations remain informative). A computed zero class in a target Hom group is a definite answer for that selected operation. For example, an explicit z-divisible Rees cocycle computes a vanishing specialization; a nonzero obstruction to a kernel comparison prevents the corresponding intertwiner; and failure of a required properness, smoothness, or coefficient-kernel hypothesis prevents applying the smooth-purity detection theorem to the selected Higgs chart. These outcomes have different mathematical meanings. In particular, a cone cannot be used as an “interchange defect” until both comparison functors and the actual natural transformation between them have been constructed.

9.5. The Atomic F-Bundle Comparison and the Singular-to-Smooth Question

Let B F be the chosen non-archimedean analytic KKPY base. The category Fbun k ( B F ) has actual horizontal morphisms and a strict exact structure; its atomic factors are selected objects satisfying additional spectral and equivariance hypotheses. The global “chemical formula” of KKPY is not itself a category of extensions. Definition 8 defines spectrally marked representatives, and Proposition 56 determines the behavior of strict extensions under compatible idempotents.
Independently, Theorem 6 constructs the exact functor
R u : FilFlat k GT ( B ) ⟶ Fbun k for ( B )
on a fixed smooth base, sending the Griffiths-transverse filtered flat bundle ( E , ∇ , F • ) to its Rees lattice with its product meromorphic connection. The functor is constructed on morphisms and preserves strictly exact sequences. For this particular unshifted construction, the Euler operator κ R u on the u = 0 fiber vanishes (Proposition 57). Exponential rank-one shifts change that operator by a scalar, as recorded in Lemma 22; the calculation is not a universal nonexistence theorem for comparisons with quantum F-bundles.
The filtered D -module functor of Section 4 and the functor (242) have different domains. In particular, the good filtered holonomic module on an ambient fourfold is not automatically a filtered flat bundle on the KKPY deformation base. Nor is there an automatic passage from a formal k-valued F-bundle to the chosen analytic, non-archimedean quantum bundle without compatible coefficient and base maps. A smooth-family Gauss–Manin construction may supply an object of the filtered-flat source on a smooth locus, but it does not by itself carry the punctual extension on the singular central fiber. The support obstruction of Proposition 60 excludes an identification with an ambient nonsingular flat bundle; it does not exclude an integrable Rees-module category with singular support. The formal z-grading and the KKPY u-connection require a verified Leibniz and integrability comparison before identification. For a non-equivariant Calabi–Yau fiber, the standard graded small-quantum Euler vector field vanishes throughout the ordinary small-quantum locus, so the particular zero-Euler mismatch disappears there; neither an atomic isomorphism nor an extension-preserving functor follows (Remark 61).
Proposition 75
(Smooth-locus factorization cannot detect the attachment). Let j : U = X 0 ∖ Σ ↪ X 0 and let T : D b MHM ( U ) → E be any exact functor to a stable category. For Φ = T ∘ j * one has
Φ ( ϵ Σ ) = 0 .
If an independently specified target atomic extension has nonzero connecting arrow ϵ Q , no endpoint isomorphisms can identify it with Φ ( ϵ Σ ) by a marked comparison square. A comparison detecting the punctual class must therefore use data that do not factor solely through j * .
Proof. 
The object R Σ H is supported on the finite closed set Σ , so j * R Σ H = 0 . Therefore j * ( ϵ Σ ) is a morphism from zero to j * I H [ 1 ] , hence is zero. Every functor sends the zero morphism to zero, giving (243). If endpoint isomorphisms identified the two arrows, then Proposition 73 would imply ϵ Q = 0 , contrary to the assumption. The final statement is the contrapositive of the factorization result; it does not assert that a particular nearby-cycle or quantum construction will succeed. □
For a specific smoothing π and a fixed quantum F-bundle, the positive comparison problem has the following exact form. Construct a geometrically defined source category C π containing the triangle (228), and construct an exact functor
R at : C π ⟶ D ex b Fbun k ( B F )
with verified action on morphisms and the necessary parameter-base comparison. For an independently defined target triangle A I → A P → A R → ϵ Q A I [ 1 ] , construct endpoint isomorphisms and verify the single marked equation
η I [ 1 ] ∘ R at ( ϵ Σ ) = ϵ Q ∘ η R .
This equation is not an asserted theorem of the present article. Its construction requires a rule incorporating the actual nearby/vanishing-cycle gluing data or another extension-sensitive operation, compatibility with the full meromorphic connection and Euler spectral operators, and the KKPY hypotheses selecting an atomic factor. It is a precise route for testing whether singular and smooth Hodge-atom data are related by morphisms rather than only by numerical invariants.
The F-bundle construction of Section 7 and the magnetic Hecke discussion of Section 8 are distinct interfaces. The former supplies an exact formal Rees construction on Griffiths-transverse filtered flat bundles, not on the singular conifold MHM extension; the latter uses the extended-Bogomolny scattering correspondence under its analytic hypotheses, not a derived MHM realization of the gauge-theoretic operation. Neither constructs the missing extension-sensitive atomic functor or a Wilson–’t Hooft duality equivalence.

9.6. From Minuscule Composition to Bivariant Correspondences

Let X and Y be separated finite-type complex algebraic schemes and let Z be a specified correspondence space:
X ← p Z → q Y .
Given a kernel K ∈ D b MHM ( Z ) and fixed boundedness, shift, twist, and star/shriek choices, the corresponding exact operation, when defined on the selected categories, is
H Z , K ( M ) = q * K ⊗ L p * M .
The essential bivariant datum is the operation together with its actual action on a marked arrow and the coherent comparison of composable correspondences. This is more information than the operator’s class on Grothendieck groups.
Definition 43
(Relation-sensitive extension datum). For the nodes of X 0 , fix the rational MHS relation object L Σ H = ker ∂ Σ H of thespecified corrected marked extension, as in Proposition 72. Under a projective small resolution with smooth source, this is D λ − 1 ker G ρ by Corollary 8. If a separate integral geometric node-relation lattice L Σ has been constructed (for example from a specified exceptional-curve, tropical, or integral chain model), arelation-sensitive comparison datumadditionally consists of a defined homomorphism
θ Σ : L Σ ⊗ Z Q ⟶ V Σ H
whose image is contained in L Σ H , together with corresponding maps after the chosen geometric correspondence and a commutative comparison square for the two global attachment boundaries. The morphism θ Σ is part of the datum; equality of the rank of L Σ with dim Q L Σ H does not define it.
Proposition 76
(Transport of a specified integral relation comparison). Suppose two such data are connected by homomorphisms u L : L Σ ⊗ Q → L Σ ′ ⊗ Q , u V : V Σ H → V Σ ′ H , and u I : I H 4 ( X 0 , Q ) → I H 4 ( X 0 ′ , Q ) , such that
u V θ Σ = θ Σ ′ u L , u I ∂ Σ H = ∂ Σ ′ H u V .
Then the image of θ Σ is carried into L Σ ′ H , compatibly with u L ; in particular the chosen integral relations remain rationally invisible under the transported global boundary. No claim about the existence of these three comparison maps follows from the proposition.
Proof. 
For ℓ ∈ L Σ ⊗ Q , compute
∂ Σ ′ H u V θ Σ ( ℓ ) = u I ∂ Σ H θ Σ ( ℓ ) = 0 .
The first equality is the second assumed commutative square; the last equality is the defining condition on θ Σ . The first assumed square identifies the same element with θ Σ ′ u L ( ℓ ) , proving compatibility. □
Remark 79
(Boundary of the bivariant continuation). For two successive finite-type correspondences satisfying the actual coefficient and exchange hypotheses, the natural equivalence (230) and its arrow square (231) may be imported into a restricted bivariant calculus. Iterated Hecke–Schober gluing, additional singular strata, and tropical-belt operations require their own correspondences, kernels, relation morphisms, and higher coherence. The computed first-jet incidence and Rees tests provide local models for such constructions, but do not determine the global operations. The same distinction applies to a comparison with KKPY quantum atoms: the constructed formal F-bundle functor is available, whereas (244) remains a distinct extension-sensitive construction problem.
The mathematical content imported by the bivariant sequel can therefore be stated concretely, in the scope of Theorem 7: a fixed conifold connecting morphism, a canonical gluing comparison, a strict filtered and Rees representative with proved nonzero conifold specialization, the framed one- and two-step Hecke geometries, a natural composition equivalence and its action on the marked arrow, the verified punctual two-step incidence, the smooth-purity nonvanishing theorem with its explicit geometric hypotheses, and the global mixed-Hodge relation exact sequences, identified with exceptional-curve class sequences when a projective small resolution exists. The corresponding singular-to-smooth atomic comparison has an identified categorical target and a proven restricted functor on filtered flat bundles; its extension-sensitive source operation is the next map to be constructed, not an identity already encoded by the notation.

Appendix A. Local shifts, Tate normalizations, and costalk complexes

The local calculations in this appendix fix the normalizations used for the attaching morphism of (15). They concern the ordinary double point and the actual mixed-Hodge-module costalk. A local extension group, a chosen generator of that group, and the globally specified variation-cone class are distinct data. In particular, the rank of the local extension group does not determine the scalar of the conifold attachment. For the corrected nonzero local class, the canonical boundary provides an independently normalized generator with costalk coordinate − 1 ; we retain the exact comparison scalar rather than silently changing the punctual endpoint marking. We use rational coefficients, the cohomological shift convention of Section 2.1, and the Hodge Tate convention wt Q H ( − m ) = 2 m .

Appendix A.1. The Ordinary Double Point and Its Link

Let Y = A C 4 have coordinates ( x , y , z , w ) and set
f = x w − y z , X = f − 1 ( 0 ) , p = ( 0 , 0 , 0 , 0 ) .
The change of complex linear coordinates identifies the quadratic form f with a nonzero multiple of z 1 2 + z 2 2 + z 3 2 + z 4 2 . Put U = X ∖ { p } , j : U ↪ X , and i : { p } ↪ X . Let B ε be a sufficiently small closed ball in Y an and let L = X ∩ ∂ B ε be the real link.
Proposition A1
(The link and the mixed Hodge cohomology of the algebraic puncture). The link L is diffeomorphic to S 2 × S 3 . For the local algebraic punctured cone U the mixed Hodge structures are
H k ( U , Q ) = Q ( 0 ) , k = 0 , Q ( − 1 ) , k = 2 , Q ( − 2 ) , k = 3 , Q ( − 3 ) , k = 5 , 0 , otherwise .
The rational cohomology of the algebraic puncture agrees with that of L as a graded vector space.
Proof. 
Consider the small resolution
ρ : X ˜ = Tot P 1 ( O P 1 ( − 1 ) ⊕ 2 ) ⟶ X .
Its exceptional fiber is E ≃ P 1 , and ρ restricts to an isomorphism X ˜ ∖ E ≃ U . The complement of the zero section in the rank-two bundle retracts onto its unit sphere bundle, with fiber S 3 over S 2 . The underlying oriented real rank-four bundle has w 2 = c 1 ( O ( − 1 ) ⊕ 2 ) mod 2 = 0 . Oriented rank-four bundles over S 2 are classified by the clutching class in π 1 ( S O ( 4 ) ) ≃ Z / 2 , detected by w 2 . Hence the sphere bundle is trivial, and its total space has the homotopy type S 2 × S 3 . The link of the conifold is the corresponding unit sphere bundle, proving the topological assertion.
For the mixed Hodge structures, use the Gysin triangle of the zero section of the algebraic rank-two bundle. Homotopy invariance gives H • ( X ˜ , Q ) = H • ( E , Q ) . The purity isomorphism for a codimension-two regular embedding identifies the relative cohomology with H • − 4 ( E , Q ) ( − 2 ) . The Euler-class homomorphism is cup product with c 2 of the bundle; it is zero because H 4 ( E , Q ) = 0 . The localization long exact sequence gives the four groups in (A2) with their indicated Tate twists. All other groups vanish. Finally, the weighted homogeneous cone contracts radially onto a punctured local neighborhood, so the same underlying rational cohomology groups compute the link. The mixed Hodge structures used here are those of the algebraic open U. □

Appendix A.2. The Localization Maps and the Gysin Long Exact Sequence

For the algebraic model in (A1), put V = O P 1 ( − 1 ) ⊕ 2 , let s : E = P 1 ↪ X ˜ = Tot ( V ) be the zero section, and let k : U = X ˜ ∖ E ↪ X ˜ be its open complement. This is an algebraic model for the punctured cone. The link L is a topological space; the mixed Hodge structures below are those on the algebraic open U. The radial homotopy equivalence with the link does not, by itself, identify the link with an algebraic mixed Hodge object.
The localization triangle and purity for the regular codimension-two embedding s give, with the indicated Tate twist,
s * Q E H ( − 2 ) [ − 4 ] ⟶ Q X ˜ H ⟶ k * Q U H → + 1 .
Here the displayed notation records the usual cohomological complexes after applying rational realization; at the level of mixed Hodge modules the standard perverse shifts are incorporated in the corresponding six-functor identifications. Applying R Γ ( X ˜ , − ) yields the actual Gysin long exact sequence
… ⟶ H k − 4 ( E ) ( − 2 ) → s * H k ( X ˜ ) → k * H k ( U ) → ∂ G , k H k − 3 ( E ) ( − 2 ) → s * H k + 1 ( X ˜ ) ⟶ … .
The first s * is the Gysin map; ∂ G , k is the connecting morphism induced by the third arrow of the localization triangle, not an independently selected generator of a Hom space. The zero-section homotopy equivalence identifies H k ( X ˜ ) with H k ( E ) , and under this identification s * s * is cup product with c 2 ( V ) , which vanishes since H 4 ( E ) = 0 .
Proposition A2
(The four nonzero groups and their actual maps). In the long exact sequence (A4), the following maps are isomorphisms of rational mixed Hodge structures:
H 0 ( X ˜ ) → k * H 0 ( U ) , H 2 ( X ˜ ) → k * H 2 ( U ) , H 3 ( U ) → ∂ G , 3 H 0 ( E ) ( − 2 ) , H 5 ( U ) → ∂ G , 5 H 2 ( E ) ( − 2 ) .
All other cohomology groups of U vanish.
Proof. 
For k = 0 , 2 both neighboring support groups in (A4) vanish; the restriction maps are therefore isomorphisms. For k = 3 its relevant portion is
0 = H 3 ( X ˜ ) ⟶ H 3 ( U ) → ∂ G , 3 H 0 ( E ) ( − 2 ) → s * H 4 ( X ˜ ) = 0 ,
so ∂ G , 3 is an isomorphism. For k = 5 it is
0 = H 5 ( X ˜ ) ⟶ H 5 ( U ) → ∂ G , 5 H 2 ( E ) ( − 2 ) → s * H 6 ( X ˜ ) = 0 .
For the other degrees the terms in the sequence vanish because H k ( E ) is nonzero only for k = 0 , 2 . Each map used is a morphism in the appropriate derived Hodge setting; hence the isomorphisms retain the Tate twists. □

Appendix A.3. Intersection-complex Stalk and Costalk

Let I H = IC X H be the pure polarizable Hodge module of weight three with rat ( I H ) = j ! * Q U [ 3 ] . We use the standard cohomological t-structure on D b MHM ( pt ) ≃ D b MHS Q for the following tables.
Proposition A3
(Stalk and costalk at the node). With the above normalization, the nonzero cohomology objects of the stalk and costalk are
H k ( i * I H ) = Q H , k = − 3 , Q H ( − 1 ) , k = − 1 , 0 , otherwise ,
H k ( i ! I H ) = Q H ( − 2 ) , k = 1 , Q H ( − 3 ) , k = 3 , 0 , otherwise .
These are statements about cohomology objects; no canonical splitting of the entire costalk complex is asserted.
Proof. 
For an isolated singularity of complex dimension three, the intersection complex on the closed cone is obtained from the intermediate extension of Q U [ 3 ] . Its local stalk can be computed by the Deligne truncation of R j * Q U [ 3 ] in ordinary cohomological degrees at most − 1 . Since H k ( i * R j * Q U [ 3 ] ) = H k + 3 ( L , Q ) , Proposition A1 gives the nonzero terms in degrees − 3 and − 1 , with Tate types Q ( 0 ) and Q ( − 1 ) . This proves (A6).
The pure intersection Hodge module is self-dual with its dimension normalization:
D X I H ≃ I H ( 3 ) .
Verdier duality commutes with the exchange of i * and i ! . Consequently
i ! I H ≃ D pt ( i * I H ) ( − 3 ) .
On a point, duality reverses cohomological degrees and sends Q H ( − r ) to Q H ( r ) . The terms in degrees − 3 , − 1 therefore give terms in degrees 3 , 1 , respectively; twisting by ( − 3 ) gives Q H ( − 3 ) and Q H ( − 2 ) . This proves (A7). □
Corollary A1
(The local forward-extension line). The adjunction i * ⊣ i ! induces a natural isomorphism
Hom D b MHM ( X ) ( i * Q H ( − 2 ) , I H [ 1 ] ) ≃ Hom D b MHS ( Q H ( − 2 ) , i ! I H [ 1 ] ) ≃ Hom MHS ( Q H ( − 2 ) , Q H ( − 2 ) ) ≃ Q .
Proof. 
The first isomorphism is derived adjunction. By (A7), the shifted complex i ! I H [ 1 ] has cohomology in degrees 0 and 2, with degree-zero term Q H ( − 2 ) . Its canonical truncation triangle is
τ ≤ 0 ( i ! I H [ 1 ] ) ⟶ i ! I H [ 1 ] ⟶ τ ≥ 1 ( i ! I H [ 1 ] ) ⟶ .
For an object of the heart MHS, there are no morphisms into an object of D ≥ 1 , nor into its desuspension in D ≥ 0 from the relevant negative side of the t-structure. More explicitly, the hyper-Ext spectral sequence for RHom ( Q H ( − 2 ) , i ! I H [ 1 ] ) has terms
E 2 a , b = Ext MHS a ( Q H ( − 2 ) , H b ( i ! I H [ 1 ] ) ) , a ≥ 0 .
Total degree zero requires ( a , b ) = ( 0 , 0 ) because b ≥ 0 . Thus the degree-zero Hom group identifies with the Hom group of the zero-th cohomology object. The endomorphisms of the one-dimensional Tate structure Q H ( − 2 ) are Q . □
Remark A1
(Other singularities and the global extension). Corollary A1 is a calculation for the ordinary double point with the stated coefficients. It is not an assertion about arbitrary singularities. For multiple nodes the local summands must be compared with the actual global attaching arrow; relations among them are not determined by the direct sum of the local one-dimensional Hom spaces.

Appendix A.4. The Costalk Long Exact Sequence of the Corrected Middle Term

We now compute the cohomology of the given middle term P H as far as it is determined by the local scalar of the specified extension; this is different from computing the scalar itself. Fix the adjunction isomorphism ϑ of (A8) and write
α p = H 0 ( i ! ϵ p ) : Q H ( − 2 ) ⟶ H 1 ( i ! I H ) = Q H ( − 2 ) , α p = c p id .
The source of α p is identified by the unit Q H ( − 2 ) → ∼ i ! i * Q H ( − 2 ) ; it is not the ordinary stalk map i * ϵ p .
Proposition A4
(Costalk and stalk groups of the middle term). Under the fixed local normalization, the long exact costalk sequence has the five-term portion
0 ⟶ H 0 ( i ! P H ) ⟶ Q H ( − 2 ) → c p id Q H ( − 2 ) ⟶ H 1 ( i ! P H ) ⟶ 0 .
Moreover H 3 ( i ! P H ) = Q H ( − 3 ) and H k ( i ! P H ) = 0 for k ∉ { 0 , 1 , 3 } . The stalk groups are
H k ( i * P H ) = Q H , k = − 3 , Q H ( − 1 ) , k = − 1 , Q H ( − 2 ) , k = 0 , 0 , otherwise .
In particular, c p ≠ 0 forces H 0 ( i ! P H ) = H 1 ( i ! P H ) = 0 ; when c p = 0 , both groups are Q H ( − 2 ) .
Proof. 
Apply i ! to I H → P H → i * Q H ( − 2 ) → ϵ p I H [ 1 ] . The punctual term has cohomology only in degree zero. From (A7), the displayed portion of its long exact cohomology sequence is exactly (A10). Its connecting map is H 0 ( i ! ϵ p ) = α p , with the same sign as the specified extension triangle. For k = 3 the sequence is 0 → Q H ( − 3 ) → H 3 ( i ! P H ) → 0 , and in the other degrees the adjacent terms vanish. Applying i * instead, use (A6) and the analogous punctual calculation. The only new group is in degree zero: 0 → H 0 ( i * P H ) → Q H ( − 2 ) → H 1 ( i * I H ) = 0 . The lower stalk groups map isomorphically from those of I H . Finally the kernel and cokernel of a nonzero scalar on a one-dimensional Tate structure are zero; for scalar zero they are the entire Tate structure. □
Remark A2
(Canonical versus corrected scalar). For the canonical recollement extension of Proposition 9, the main text identifies c can , p = − 1 under its explicit cone convention, so the two degree-zero/one costalk groups vanish. For the nonzero corrected extension write ϵ p = λ p ϵ can , p , with λ p ∈ Q × ; then c p = − λ p . Thus its degree-zero and degree-one costalk groups also vanish, even though the exact marked scalar remains undetermined. Determination of the numerical value λ p still requires the actual normalized variation-cone-to-recollement map and the endpoint comparison. The calculation above gives an independent check on that comparison; it does not assert c p = − 1 by analogy.
Proposition A5
(Vanishing under strict punctual restriction). For the local forward attaching arrow ϵ p : i * Q H ( − 2 ) → I H [ 1 ] , one has
i * ϵ p = 0 in D b MHS Q .
Consequently an algebraic source map factoring through the reduced point p pulls this arrow back to zero; any subsequent derived tensor and *- or !-pushforward operation sends that restricted arrow to zero as well. This assertion does not evaluate the arrow on a full Hecke source retaining directions transverse to p.
Proof. 
The stalk table (A6) gives cohomology of i * I H [ 1 ] only in degrees − 4 and − 2 . Since i * i * Q H ( − 2 ) = Q H ( − 2 ) , the restricted arrow belongs to
Hom D b MHS Q ( Q H ( − 2 ) , i * I H [ 1 ] ) .
In the hyper-Ext spectral sequence, total degree zero would require an Ext 4 contribution from degree − 4 or an Ext 2 contribution from degree − 2 . The category of rational mixed Hodge structures has homological dimension one, so both groups vanish [24]. Hence the displayed Hom group is zero. For a source map p Z : Z → X factoring as Z → { p } → i X , pseudofunctoriality gives p Z * ϵ p = 0 . The remaining assertions follow from functoriality of tensor products and direct images. In particular, this stalkwise vanishing does not contradict the nonzero costalk coordinate c p in (A9). □

Appendix A.5. Nearby and Vanishing Cycles; the Variation Fiber

Let f : Y → A 1 be a morphism on a smooth complex algebraic variety and let M be a specializable Hodge module. Write Ψ = ψ f , 1 H ( M ) [ − 1 ] and Φ = ϕ f , 1 H ( M ) [ − 1 ] only when the functors are initially unshifted; if they are already perverse normalized, suppress the additional shifts. Fix maps
can H : Ψ → Φ , var H : Φ → Ψ ( − 1 ) .
For the logarithmic normalization of variation, write N Ψ and N Φ for the nilpotent monodromy logarithms, with their Tate-twisted targets.
Proposition A6
(Monodromy identities and the variation cone). The compositions satisfy
var H can H = N Ψ , can H ( − 1 ) var H = N Φ .
Put
V f ( M ) = Cone ( var H ) [ − 1 ] = fib ( var H ) .
There is a distinguished triangle
V f ( M ) ⟶ Φ → var H Ψ ( − 1 ) ⟶ V f ( M ) [ 1 ] .
If Φ and Ψ ( − 1 ) lie in the perverse heart, then
H 0 ( V f ( M ) ) = ker ( var H ) , H 1 ( V f ( M ) ) = coker ( var H ) ,
and all other perverse cohomology objects vanish.
Proof. 
The identities are Saito’s normalized canonical and logarithmic variation relations, with the indicated Tate-twisted targets. The cone convention of Section 2.1 gives the fiber triangle (A15). Applying perverse cohomology to that triangle yields the exact sequence
0 → H 0 ( V f ( M ) ) → Φ → var H Ψ ( − 1 ) → H 1 ( V f ( M ) ) → 0 ,
with the claimed vanishing outside degrees zero and one. □
Remark A3
(Comparison with the corrected middle term). The equality P H ≃ V f ( M ) cannot be inferred from the definition of the fiber alone. It requires the truncation, recollement, and gluing comparison of the corrected construction [2]. Replacing logarithmic variation by the ordinary topological variation simultaneously changes the monodromy identities from N to T u − 1 ; the two conventions must not be combined without the appropriate power-series factor.

Appendix A.6. The Canonical Boundary as a Morphism in Recollement

For a finite node set let i : Σ ↪ X 0 , j : U ↪ X 0 and I H = IC X 0 H . Put A = i * i ! I H and B = j * j * I H in D b MHM ( X 0 ) . The counit i * i ! I H → I H and the adjunction unit I H → j * j * I H form the localization triangle
A ⟶ I H ⟶ B → ∂ A [ 1 ] .
The point costalk computation gives H 0 ( A ) = 0 and H 1 ( A ) = R Σ H ; the additional H 3 ( A ) does not enter degree zero. The long exact sequence begins
0 = H 0 ( A ) ⟶ I H → H 0 ( unit ) H 0 ( B ) → H 0 ( ∂ ) H 1 ( A ) = R Σ H ⟶ H 1 ( I H ) = 0 .
Thus P can H = H 0 ( B ) is the middle term of a short exact sequence with the constructed maps u can and v can above.
Proposition A7
(A checkable marked comparison). Let 0 → I H → u P H → v R Σ H → 0 be the corrected extension. Its adjunction unit induces a map a P : P H → P can H extending id I H ; write h R : R Σ H → R Σ H for the induced quotient map. Then
ϵ can ∘ h R = ϵ Σ .
After the fixed identifications of the punctual endpoints, a P is an isomorphism if and only if h R is, whereas it is an isomorphism ofmarkedshort exact sequences with fixed quotient marking if and only if h R = id .
Proof. 
Since j * u is an isomorphism, the units of the adjunction j * ⊣ j * determine the composite P H → j * j * P H → ∼ j * j * I H . Take its zeroth cohomology (both source and target are in the MHM heart after taking H 0 ) to obtain a P . Functoriality of the adjunction unit shows that the restriction of a P along u is u can ; therefore it induces h R on the cokernels. We have the commutative diagram
0 → I H → u P H → v R Σ H → 0 id ↓ id a P ↓ a P h R ↓ h R 0 → I H → u can P can H → v can R Σ H → 0 .
It is a morphism of short exact sequences, so naturality of the Yoneda boundary gives (A18). By the short five lemma a P is an isomorphism when h R is; conversely, if a P is an isomorphism its quotient map h R is as well. Preserving the chosen quotient marking means exactly h R = id , which is stronger than invertibility. No scalar has been chosen after this map was constructed. □

Appendix A.7. Exceptional-curve Normalization and the Remaining Scalar

For the small resolution ρ : X ˜ → X of Proposition A1, let E = ρ − 1 ( p ) ≃ P 1 . Let h = c 1 ( O E ( 1 ) ) ∈ H 2 ( E , Q ( 1 ) ) and fix the complex orientation so that
∫ E h = 1 .
This is a cohomological class; it is not identified with the homology fundamental class without Poincare duality.
Definition A1
(Costalk-normalized extension coordinate). Let ϑ be the isomorphism in (A8). For an arrow e : i * Q H ( − 2 ) → I H [ 1 ] , define λ ( e ) ∈ Q by
ϑ ( e ) = λ ( e ) id Q H ( − 2 ) .
Proposition A8
(Normalized comparison criterion). Suppose a geometrically constructed exceptional-curve boundary map e E : i * Q H ( − 2 ) → I H [ 1 ] is given, with its orientation and Hodge normalization specified. Then e E = e if and only if λ ( e E ) = λ ( e ) . In particular, the sign of the comparison with the variation-cone class is determined by the costalk map i ! ( e ) under (A8), not by the abstract dimension of the extension group.
Proof. 
By Corollary A1, ϑ is an isomorphism to a one-dimensional rational vector space, hence is injective. Therefore e E = e is equivalent to ϑ ( e E ) = ϑ ( e ) , which is equivalent to the equality of their rational coordinates. The coordinate is defined using the actual costalk morphism, so the criterion retains the specified maps rather than selecting an extension by its desired image. □
Remark A4
(What the local topology does not determine). The normalization (A19) fixes the orientation of h and the resulting trace map. The canonical geometric Gysin arrow is constructed in Proposition A9, with costalk coordinate + 1 ; the canonical recollement boundary therefore has the negative coordinate. This does not by itself compute λ ( ϵ p ) for the globally specified variation-cone extension: that value requires the actual normalized variation-cone comparison from the corrected conifold construction. In the coordinate of Definition A1, λ ( ϵ can , p ) = − 1 and λ ( ϵ p ) = − λ p for ϵ p = λ p ϵ can , p . Thus the nonzero condition is known, while the further assertion λ p = 1 (or even λ p = ± 1 ) has not been established by this topological trace calculation.

Appendix A.8. Global Exceptional Curves and the Canonical boundary–Gysin Identification

Let X 0 be a projective nodal complex threefold, let a : X 0 → pt be its structure morphism, and assume that ρ : X ˜ 0 → X 0 is a projective small resolution with X ˜ 0 smooth and C p = ρ − 1 ( p ) ≃ P 1 for every p ∈ Σ . Write s p : C p ↪ X ˜ 0 , b p : C p → { p } , and i p : { p } ↪ X 0 . The square ρ s p = i p b p is Cartesian. Smallness and the decomposition theorem supply the canonical small-map identification θ ρ : I H → ∼ R ρ * Q X ˜ 0 H [ 3 ] when the restriction over U is normalized by the identity on Q U H [ 3 ] . We use the complex orientation to normalize purity for the regular codimension-two embedding s p ; in particular, s p ! Q X ˜ 0 H ≃ Q C p H ( − 2 ) [ − 4 ] .
Proposition A9
(Construction of the exceptional-curve Gysin arrow). The following composite defines a morphism of derived mixed Hodge modules, with no auxiliary choice of a generator:
γ p : i p * Q p H ( − 2 ) → i p * η b p i p * R b p * Q C p H ( − 2 ) ≃ R ρ * s p * Q C p H ( − 2 ) → R ρ * ( Gys s p [ 4 ] ) R ρ * Q X ˜ 0 H [ 4 ] → θ ρ − 1 [ 1 ] I H [ 1 ] .
Here η b p : Q p H ( − 2 ) → R b p * Q C p H ( − 2 ) is the adjunction unit and Gys s p : s p * Q C p H ( − 2 ) [ − 4 ] → Q X ˜ 0 H is the purity counit. Under the derived-adjunction costalk coordinate of (A8), the local arrow γ p has coordinate + 1 . Under proper global sections, its degree-zero boundary is the cohomology class PD [ C p ] in H 4 ( X ˜ 0 , Q ) , using the standard Tate marking.
Proof. 
The purity isomorphism identifies the Gysin map with the counit of ( s p * , s p ! ) . Proper extraordinary base change in the displayed Cartesian square, followed by θ ρ , gives
i p ! I H [ 1 ] ≃ i p ! R ρ * Q X ˜ 0 H [ 4 ] ≃ R b p * s p ! Q X ˜ 0 H [ 4 ] ≃ R Γ ( C p , Q C p H ( − 2 ) ) .
The mate of the composite (A20) under i p * ⊣ i p ! is precisely the unit Q p H ( − 2 ) → R b p * Q C p H ( − 2 ) . Taking degree zero sends 1 to the constant section 1 in H 0 ( C p , Q ) . Thus its costalk coordinate is + 1 , not an arbitrarily assigned nonzero scalar. On applying proper global sections of X ˜ 0 to the purity counit, the image of this unit is the regular-embedding cycle-class morphism; by the normalization of the complex orientation, this maps 1 ∈ H 0 ( C p , Q ) ( − 2 ) to s p * ( 1 ) = PD [ C p ] ∈ H 4 ( X ˜ 0 , Q ) . These identifications are natural for the selected resolution and purity data. □
Theorem A1
(Canonical global boundary and exceptional-curve relations). With the hypotheses and normalizations above, put V Σ H = ⨁ p ∈ Σ Q H ( − 2 ) , G ρ : V Σ H → H 4 ( X ˜ 0 , Q ) , G ρ ( e p ) = PD [ C p ] , and let D λ be the rational diagonal marking from (87). Then
θ ρ ∂ Σ , can H = − G ρ , θ ρ ∂ Σ , corr H = − G ρ D λ , L Σ , corr H = D λ − 1 ker G ρ .
Here the sign is the rotation sign of Proposition 9, applied after the fixed purity normalization; it is not a sign assigned from the orientation of the exceptional curve. The formula is an identity of rational Hodge-structure morphisms and relation subspaces. It does not identify integral relation lattices or compute the independently marked variation-cone scalars λ p .
Proof. 
The canonical boundary of the recollement extension is ϵ can , p = − β can , p [ 1 ] with costalk coordinate − 1 by Proposition 9. By Proposition A9, γ p has coordinate + 1 . The local derived Hom group is the one-dimensional rational line computed in Corollary A1; hence ϵ can , p = − γ p as actual morphisms, with both endpoints and their markings fixed. Proper direct image then takes each canonical local column to − PD [ C p ] , establishing the first identity without an assumed boundary–Gysin comparison. The already proved relation ∂ Σ , corr H = ∂ Σ , can H D λ gives the second. Because D λ is invertible over Q , taking kernels proves the third. No integral statement follows from this rational change of coordinates. □
Remark A5
(Scope of the geometric identification). The Gysin arrow is a constructed morphism, not an identification obtained by counting dimensions. In particular, the small-map isomorphism is normalized on the smooth locus and the purity counit uses the complex orientation; changing either marking changes the coordinate description. The conclusion requires an actual smooth projective small resolution and does not apply to an arbitrary nodal threefold merely because its nodes have local analytic small resolutions. Moreover, ϵ p = λ p ϵ can , p still leaves the corrected marked scalar λ p unresolved.

Appendix B. Rees Complexes, Derived Specialization, and Hom Calculations

This appendix isolates the exact algebra behind Section 4. The first calculation takes place in a smooth ambient fourfold and fixes the underlying differential-module presentation of the punctual endpoint. The next two calculations are fully explicit strict filtered examples. They demonstrate why exactness of associated graded does not imply that a nonsplit extension remains nonsplit. For the determinantal conifold the good filtered localization model and its first Hodge levels are now calculated in Theorem 2 and Lemma 6. The resulting non-splitting is proved in Theorem 3. The algebra below remains useful for describing the actual Rees cocycle and testing arbitrary other filtered extensions.

Appendix B.1. Ambient Differential Modules and the Punctual Endpoint

Let Y = A C 4 , X = { x w − y z = 0 } ⊂ Y , and k : X ↪ Y the closed embedding. Let D Y be the sheaf of algebraic differential operators with its order filtration; we use left  D -modules. For a filtered left module ( M , F ) write
DR Y ( M ) = M → Ω Y 1 ⊗ M → … → Ω Y 4 ⊗ M [ 4 ] ,
where the unshifted de Rham complex has degrees 0 , … , 4 . Thus the shifted de Rham functor takes regular holonomic modules to the perverse normalization. A right-module convention requires tensoring with ω Y and changing the filtration indices.
Proposition A10
(Underlying punctual differential module). Let i Y : { p } ↪ Y denote the origin. The left D Y -module underlying i Y * Q H ( − 2 ) has the local presentation
δ p ≃ D Y / D Y x + D Y y + D Y z + D Y w .
As an O -module it is spanned by the derivatives ∂ x a ∂ y b ∂ z c ∂ w d δ . The Hodge filtration is the shifted direct-image filtration attached to the Tate-twisted point object, not an arbitrarily assigned order filtration.
Proof. 
The direct image of the one-dimensional left differential module on a point is generated by a delta symbol δ annihilated by the four coordinate functions. The transfer bimodule for a closed embedding identifies the resulting module with the quotient in (A23). The Poincare–Birkhoff–Witt normal form gives the indicated spanning set, with no relations among the differential monomials. Saito’s filtered direct image fixes the filtration on the transfer module; the Tate twist changes its filtration indices while leaving the underlying D Y -module unchanged. Therefore the displayed quotient is a presentation of the underlying module but is not yet a complete filtered presentation without those indices. □
Remark A6
(The middle module is additional data). Let M I , M P , M R be the filtered components on Y of I H , P H , R Σ H , transported by Kashiwara equivalence. The exact sequence of MHM supplies a strictly exact filtered sequence
0 → ( M I , F ) → ( M P , F ) → ( M R , F ) → 0 .
The quotient presentation (A23) does not determine the filtration on M P . That filtration has been obtained for the canonical extension in the localization model of Section 4, and for the nonzero corrected class by scalar-equivalence of its marked endpoints. The punctual presentation alone would not have supplied this result.
Remark A7
(Localization proof and its precise filtration input). For the four-variable quadratic node g = x w − y z , the minimal exponent is α ˜ g = 2 . The Hodge-ideal criterion k + 1 ≤ α ˜ g for the triviality of the relevant Hodge ideal applies for k = 0 , 1 , giving the pole-order description of the two filtration levels used in Lemma 6 [20]. With the fixed residue twist ( 1 ) , gr 2 F P can , p ≃ O Y , p / ( g ) ; this ring has no nonzero element annihilated by the maximal ideal. That is the essential obstruction in Theorem 3. It should not be confused with mere z-torsion-freeness of a Rees module.

Appendix B.2. A Graded Algebraic Model for Good Rees Extensions

Set A = C [ u ] with the degree filtration F p A = { f : deg ( f ) ≤ p } for p ≥ 0 , and F p A = 0 for p < 0 . Its Rees ring is
R = R F A ≃ C [ z , U ] , U = u z ,
where z has filtration degree one in the Rees convention and U has the corresponding homogeneous degree one. We use R 〈 r 〉 for the grading shift that places its chosen free generator in degree r. In the examples below the degree assignments on presentations are chosen so that every differential has degree zero.
Proposition A11
(A surviving extension). The strictly homogeneous sequence
0 ⟶ R 〈 1 〉 → U R ⟶ R / ( U ) ⟶ 0
is nonsplit as a sequence of R-modules. Its specialization at z = 1 is the nonsplit sequence 0 → A → u A → A / ( u ) → 0 , with the indicated filtration shift. At z = 0 it becomes
0 → C [ U ] 〈 1 〉 → U C [ U ] → C [ U ] / ( U ) → 0 ,
which is also nonsplit. Each of the three terms in (A24) is z-torsion-free.
Proof. 
Multiplication by the indeterminate U is injective and has cokernel R / ( U ) . A splitting would make R / ( U ) projective as an R-module, which is impossible: applying − ⊗ R R / ( U ) to the sequence leaves the nonzero group Tor 1 R ( R / ( U ) , R / ( U ) ) ≃ R / ( U ) . Because U and z are algebraically independent, R, R 〈 1 〉 , and R / ( U ) ≃ C [ z ] are z-torsion-free. Hence derived z-specialization agrees with ordinary quotient by z on each term and preserves exactness. The two specialized sequences are again multiplication by a nonunit parameter in a polynomial ring, and the same Tor argument proves nonsplitting. □
Proposition A12
(A nonsplit extension killed by specialization). In R = C [ z , U ] define
E = ( R a ⊕ R e ) / ( U e − z a ) ,
with the degrees chosen so that U e and z a are homogeneous of the same degree. There is a short exact sequence
0 ⟶ R a ⟶ E ⟶ R / ( U ) e ¯ ⟶ 0 .
Its extension class corresponds to z under
Ext R 1 ( R / ( U ) , R ) ≃ R / ( U ) = C [ z ] .
Thus it is nonsplit over R, becomes nonsplit at z = 1 , and splits at z = 0 . All terms in (A26) are z-torsion-free, so the sequence is strict in the Rees sense.
Proof. 
The map R a → E is injective: if r a lies in the submodule generated by U e − z a , comparison of the e-coefficients gives U q = 0 , hence q = 0 and r = 0 . The quotient of E by R a is generated by e ¯ with the single relation U e ¯ = 0 . This proves exactness.
A free resolution of the quotient is
0 ⟶ R → U R ⟶ R / ( U ) ⟶ 0 ,
with suitable grading shifts. Applying Hom R ( − , R ) gives the two-term complex R → U R in degrees zero and one. Therefore Ext R 0 ( R / ( U ) , R ) = 0 and Ext R 1 ( R / ( U ) , R ) = R / ( U ) , with no higher Ext. Lifting the generator e ¯ to e gives the relation U e = z a ; the associated connecting class is exactly [ z ] modulo U.
At z = 1 the relation reads U e = a , so E / ( z − 1 ) E is free on e and the injection of R a becomes multiplication by U, which is nonsplit. At z = 0 the relation becomes U e = 0 , so
E / z E ≃ C [ U ] a ⊕ C [ U ] / ( U ) e ,
and the specialized extension splits. Finally, E has the free resolution
0 ⟶ R → ( − z , U ) R 2 ⟶ E ⟶ 0 .
The vector ( − z , U ) remains injective after specializing at z = 0 . It follows directly that E has no z-torsion and that Tor 1 C [ z ] ( E , C ) = 0 . The quotient R / ( U ) = C [ z ] is also z-flat. This proves strictness and the remaining claims. □

Appendix B.3. Derived Specialization and the Hom Kernel

For a C [ z ] -linear dg or stable enhancement D R of suitable graded R-modules, let sp 0 ( − ) = − ⊗ C [ z ] L C with z acting by zero, and let D 0 be the corresponding graded R / ( z ) category. For objects M , N such that the operation is defined, functoriality gives a map
Hom D R ( M , N [ 1 ] ) ⟶ Hom D 0 ( sp 0 M , sp 0 N [ 1 ] ) .
Proposition A13
(Explicit specialization kernel). For M = R / ( U ) and N = R , with graded shifts accounted for, the ungraded total Ext map underlying (A28) is
C [ z ] ⟶ C , f ( z ) ⟼ f ( 0 ) .
Its kernel is z C [ z ] . The class of Proposition A11 is represented by the unit 1, whereas the class of Proposition A12 is represented by z. The former survives and the latter vanishes.
Proof. 
Apply Hom to the two-term free resolution of R / ( U ) used in Proposition A12. Over R the resulting cochain complex is
R → U R ,
concentrated in degrees zero and one. Its only cohomology is Ext R 1 ( R / ( U ) , R ) = R / ( U ) = C [ z ] . After derived specialization, the resolution becomes
C [ U ] → U C [ U ] ,
again concentrated in degrees zero and one, with degree-one cohomology C . The natural map of cochain complexes is reduction modulo z in each degree; hence it induces evaluation f ( z ) ↦ f ( 0 ) on degree-one cohomology. Its kernel is the principal ideal z C [ z ] . The presentations (A24) and (A26) give the classes 1 and z, respectively. The vanishing and survival assertions follow. □
Remark A8
(Cohomological degrees and Tor). The free resolution of R / ( U ) has homological degrees − 1 , 0 ; its derived Hom has cohomological degrees 0 , 1 , with H 0 = 0 and H 1 = C [ z ] . Specialization preserves this amplitude, with H 0 = 0 and H 1 = C . All modules displayed in the two strict examples are C [ z ] -torsion-free, hence flat over the principal ideal domain C [ z ] ; their higher Tor j C [ z ] ( − , C ) vanish for j > 0 . Failure of extension survival is thus compatible with vanishing Tor of all three sequence terms.

Appendix B.4. The Four-Coordinate Koszul Computation for the Punctual Rees Quotient

This is the local calculation promised in Section 4.9: it uses the actual punctual endpoint, not a free polynomial toy quotient. Let D be the fourth Weyl algebra in ordered coordinates x 1 = x , x 2 = y , x 3 = z , x 4 = w . Write R D for its order-Rees algebra, with ξ i = z ∂ x i , so [ ξ i , x j ] = δ i j z ; the x i have degree zero. For the left filtered punctual module underlying i o * Q H ( − 2 ) , Section 4 calculates the Rees generator degree as 2, after the codimension-four direct-image shift and the ( − 2 ) Tate twist. The quotient is
R ˜ o ≃ ( R D / ∑ i = 1 4 R D x i ) 〈 2 〉 .
Since right multiplication by the commuting x i is R D -linear on a free left module, its Koszul resolution has terms
0 ⟶ ( R D 〈 2 〉 ) 1 ⟶ ( R D 〈 2 〉 ) 4 ⟶ ( R D 〈 2 〉 ) 6 ⟶ ( R D 〈 2 〉 ) 4 ⟶ R D 〈 2 〉 ⟶ R ˜ o ⟶ 0 .
The maps are given by the alternating right-multiplication formula (115); this order is essential for left D-modules.
Proposition A14
(Koszul cocycles, boundaries, and derived reduction). Let 0 → A ˜ o → a R P ˜ o → R ˜ o → 0 be the strict Rees sequence of the corrected conifold extension on this chart. Choose a homogeneous lift m ∈ ( P ˜ o ) 2 of the fixed punctual generator, and put a i = a R − 1 ( x i m ) . Then
C 0 = ( A ˜ o ) 2 , C 1 = ( A ˜ o ) 2 4 , C 2 = ( A ˜ o ) 2 6 , d C 0 ( b ) = ( x 1 b , x 2 b , x 3 b , x 4 b ) , ( d C 1 a ) i j = x i a j − x j a i ( i < j ) .
The class of a = ( a 1 , … , a 4 ) in H 1 ( C • ) represents the specified conifold Rees extension. After z = 0 its class vanishes precisely when
a ¯ i = x i b ¯ ( i = 1 , 2 , 3 , 4 ) for some b ¯ ∈ ( gr F M I ) 2 .
Proof. 
The short exact sequence makes each x i m an element of ker ( P ˜ o → R ˜ o ) = a R ( A ˜ o ) , so each a i exists uniquely. Commutativity of the coordinate operators gives a R ( x i a j − x j a i ) = x i x j m − x j x i m = 0 , whence d C 1 a = 0 . If m is replaced by m + a R ( b ) , each a i changes to a i + x i b , exactly d C 0 ( b ) . Thus the cocycle class does not depend on the lift. The connecting morphism is the usual obstruction to lifting the augmentation of the resolution to the middle term; its first obstruction is the displayed tuple, and its higher Koszul relations are precisely the cocycle equations. The free resolution (A31) stays exact after z = 0 : the commutator relations become zero and x 1 , … , x 4 remain a regular sequence in the polynomial associated-graded Weyl ring. Derived base change is therefore represented by reducing the cochain maps modulo z. The reduced class is zero exactly when the reduced tuple is in the image of d C 0 , giving the four simultaneous equations (A33). □
Remark A9
(Relation to the completed conifold non-splitting theorem). The proposition computes the resolution and the exact coboundary problem; the four component symbols by themselves do not determine the extension cocycle. Nevertheless R3 is no longer open: Theorem 2 identifies the canonical middle filtered module with the localization quotient, and Theorem 3 proves that its associated-graded extension does not split. Thus the simultaneous system (A33) has no solution for the canonical cocycle. Rescaling the punctual endpoint by a nonzero rational number gives the same conclusion for the corrected local class, independently of the unresolvedmarkedscalar. This is a statement about the actual conifold Hodge filtration; it is not inferred from characteristic support alone.

Appendix B.5. The Canonical Localization Model and the Graded Obstruction

On the ambient smooth fourfold with g = x w − y z , let j Y : Y ∖ X ↪ Y . The canonical extension has, under the supported Hodge-module identification, the left filtered differential-module representative
( O Y [ g − 1 ] / O Y , F ) ( 1 ) ,
by Theorem 2. With the normalization of Section 4, its first two shifted filtration levels give
gr 2 F P can , o ≃ O Y , o / ( g ) , gr 2 F R o ≃ C o .
Here the punctual graded generator is annihilated by the maximal ideal ( x , y , z , w ) . If the associated-graded extension split, its section on this generator would give a nonzero element t ∈ O Y , o / ( g ) annihilated by each coordinate. Since g is irreducible, the latter ring is an integral domain, and x t = 0 forces t = 0 , a contradiction. Therefore the associated-graded extension is nonsplit and its connecting arrow is nonzero. The same holds after any nonzero rational rescaling of its source marking, including the corrected conifold class.
Remark A10
(How the localization proof meets the Koszul test). The degree-two punctual generator is the augmentation generator in the four-coordinate Koszul resolution (A31). A solution of (A33) would make the reduced extension a boundary and hence supply the forbidden graded splitting. The localization argument rules this out without selecting an arbitrary generator of the extension space or assuming that specialization detects every filtered extension.

Appendix B.6. The Bockstein Long Exact Sequence on Extension Morphisms

Let A ˜ , C ˜ be Rees modules in the graded derived category of Section 4.6. Multiplication by the central parameter z is homogeneous of degree zero as a map A ˜ ( − 1 ) → A ˜ . Its cone is the derived special fiber. The associated triangle is
A ˜ ( − 1 ) → z A ˜ ⟶ sp 0 A ˜ ⟶ A ˜ ( − 1 ) [ 1 ] .
Applying Hom ( C ˜ , − ) and using the derived extension-of-scalars adjunction gives the following long exact segment:
Hom ( C ˜ , A ˜ ( − 1 ) [ 1 ] ) → z ∘ − Hom ( C ˜ , A ˜ [ 1 ] ) → sp 0 Hom ( sp 0 C ˜ , ( sp 0 A ˜ ) [ 1 ] ) → δ z Hom ( C ˜ , A ˜ ( − 1 ) [ 2 ] ) → z ∘ − Hom ( C ˜ , A ˜ [ 2 ] ) .
Here all unadorned Hom sets on the right are in the graded Rees-derived category; the middle one is over the specialized algebra. Consequently
ker sp 0 : Hom ( C ˜ , A ˜ [ 1 ] ) → Hom ( sp 0 C ˜ , ( sp 0 A ˜ ) [ 1 ] ) = im ( z ∘ − ) .
This is a complete formal criterion, while (A33) is its concrete local four-coordinate implementation. It does not identify the filtered Rees parameter z with the auxiliary variable u of Theorem 6; those belong to different constructions and have different meanings.

Appendix B.7. A Local Finite-Resolution Criterion for the Conifold

Let R = R F D Y , and let M R = R F M R , M I = R F M I . Assume a specified finite good graded locally free resolution P • → M R in the selected local model, compatible with the Rees filtration, and suppose the base-change morphism for the chosen Hom complex is defined. Set
C • = Hom R ( P • , M I ) , C 0 • = C • ⊗ C [ z ] L C .
Write ξ ∈ H 1 ( C • ) for the extension class determined by the given Saito sequence, not by an arbitrary generator.
Proposition A15
(Finite-complex survival criterion). Assume C • is a valid model for derived Hom, that its base change computes derived Hom of the specialized endpoints, and that ξ is represented by a degree-one cocycle c. Then the specialized arrow is nonzero if and only if the class of c ⊗ 1 in H 1 ( C 0 • ) is nonzero. If C • is a bounded complex of C [ z ] -flat modules and the usual base-change spectral sequence is convergent, its page is
E 2 − a , b = Tor a C [ z ] ( H b ( C • ) , C ) ⟹ H b − a ( C 0 • ) ,
where a ≥ 0 . Since C [ z ] has global dimension one, only the columns a = 0 , 1 occur.
Proof. 
The chosen representative c defines a morphism from the resolution P • to M I [ 1 ] ; applying the derived base-change functor sends it to the class represented by c ⊗ 1 . A morphism in the homotopy category is zero precisely when its class in degree-one derived Hom vanishes, proving the first claim. For the spectral sequence resolve C as a C [ z ] -module by 0 → C [ z ] → z C [ z ] → C → 0 . Tensoring this resolution with the flat complex C • gives a bounded double complex. Filtering by the resolution degree gives the Tor spectral sequence displayed above. The resolution has length one, so the stated vanishing columns follow. In this case no differential d r for r ≥ 2 has both its source and target in the nonzero columns; the remaining issue is the edge map and its possible extension filtration, not an uncomputed higher differential. □
Remark A11
(What has and has not been computed). Propositions A13 and A15 provide an algebraic specialization criterion, not an additional prerequisite to the already proved determinantal-node result. In the present case the localization identification and the domain argument of Theorem 3 prove ϵ 0 , can ≠ 0 and ϵ 0 , corr ≠ 0 . A separate explicit Koszul representative may still be useful for computational checks or other coefficient classes. No equality of corrected and canonicalmarkedarrows has been assumed.
Remark A12
(Singular-support Rees modules are not excluded). The filtered localization quotient O Y [ g − 1 ] / O Y is not the differential module of a nonsingular flat vector bundle on all of Y: its support is the proper hypersurface X. This doesnotexclude considering the object in an independently specified category of integrable modules over the Rees algebra R F D Y , where singular support is permitted. A prospective auxiliary-direction operator such as a grading action z 2 ∂ z must obey the relevant Leibniz identities, preserve the chosen Rees lattice in the required sense, and satisfy the integrability and base-comparison conditions of its target category. The non-splitting theorem at z = 0 does not supply these additional structures or an isomorphism with a KKPY quantum F-bundle.

Appendix C. Rank-Two Hecke Lattices, Kernels, And Twofold Convolution

The formulas below verify the framed G L 2 lattice geometry used in Section 5 and Section 6. They distinguish the full lattice convolution from its Higgs-invariant subspace and the latter from the corresponding operation on derived MHM. The spherical tensor-square decomposition is calculated in its own category; its action on a conifold attachment requires the separate kernel comparison specified in Section 6.6.

Appendix C.1. One-step Modifications and Changes of Frame

Let O = C [ [ w ] ] , K = C ( ( w ) ) , and Λ 0 = O e 1 ⊕ O e 2 . For a line ℓ = [ s : t ] ∈ P ( Λ 0 / w Λ 0 ) , let λ : Λ 0 → Λ 0 / w Λ 0 be reduction and set
Λ ℓ = λ − 1 ( ℓ ) .
Proposition A16
(Local charts of the minuscule Grassmannian). The lattice Λ ℓ satisfies
w Λ 0 ⊂ Λ ℓ ⊂ Λ 0 , dim C ( Λ 0 / Λ ℓ ) = 1 .
On the chart s ≠ 0 , put u = t / s . Then
Λ ℓ = O ( e 1 + u e 2 ) ⊕ O ( w e 2 ) .
On the chart t ≠ 0 , put v = s / t . Then
Λ ℓ = O ( v e 1 + e 2 ) ⊕ O ( w e 1 ) .
On the overlap u v = 1 , and the two bases differ by an element of G L 2 ( O ) .
Proof. 
The inverse image of a one-dimensional subspace modulo w is a rank-two lattice containing w Λ 0 . The quotient is the one-dimensional complementary quotient of Λ 0 / w Λ 0 , proving colength one. On s ≠ 0 the line is spanned by ( 1 , u ) , so a vector a e 1 + b e 2 lies in Λ ℓ precisely when b ≡ u a ( mod w ) . Therefore it is an O -linear combination of e 1 + u e 2 and w e 2 . The other chart is analogous. On the overlap v = u − 1 and
v e 1 + e 2 = v ( e 1 + u e 2 ) , w e 1 = w ( e 1 + u e 2 ) − u ( w e 2 ) .
The change-of-basis matrix has determinant a unit of O , as required. □
Remark A13
(Frame covariance). For g ∈ G L 2 ( O ) , the lattice changes by Λ ℓ ↦ g Λ ℓ , while its parameter changes by ℓ ↦ g ¯ ℓ , where g ¯ ∈ G L 2 ( C ) is reduction modulo w. An object constructed on a framed chart descends only after the corresponding equivariance and cocycle data have been verified.

Appendix C.2. The Full Incidence Versus the Kernel-Line Small Resolution

The determinantal conifold has two different incidence constructions that must not be exchanged when applying a pushforward. Write A = a b c d and X det = { a d − b c = 0 } ⊂ Mat 2 ( C ) . Let
I 0 = { ( A , ℓ ) : det A = 0 , A ℓ ⊆ ℓ } , J ker = { ( A , ℓ ) : A ℓ = 0 } .
On the chart ℓ = C ( 1 , u ) the invariant-line equation is c + ( d − a ) u − b u 2 = 0 . Set ξ = a + b u and ζ = d − b u . The determinant then factors as det A = ξ ζ . Therefore, on this chart,
I 0 = V ( ξ ζ ) , J ker = V ( ξ ) ,
where I 0 generally has two components, including the quotient-eigenvalue component V ( ζ ) .
Proposition A17
(Geometry and fibers of the two projections). The projection J ker → X det is a small resolution, with total space Tot P 1 ( O ( − 1 ) ⊕ 2 ) ; over each nonzero rank-one matrix its fiber is the single line ker A , and over A = 0 its fiber is P 1 . By contrast, I 0 → X det has two invariant lines over a generic rank-one matrix with distinct eigenvalues 0 and tr A , while its fiber over zero is also P 1 . Consequently equality of the exceptional fibers is insufficient to identify the direct-image functors.
Proof. 
For a fixed line ℓ ⊂ C 2 , an endomorphism annihilating ℓ is a linear map from the quotient to C 2 . Over P 1 these maps form the rank-two bundle
Hom ( Q , O ⊕ 2 ) ≃ Q * ⊗ C 2 ≃ O ( − 1 ) ⊕ 2 , Q ≃ O ( 1 ) .
Every nonzero rank-one A ∈ X det has a unique kernel line, so the map is an isomorphism over X det ∖ { 0 } ; over zero the kernel condition is automatic and its fiber is P 1 . The exceptional locus has codimension two in the smooth threefold total space, hence the resolution is small. For I 0 , a generic rank-one A has two distinct eigenspaces, each an invariant line; for A = 0 every line is invariant. The chart calculation (A41) records these two geometric branches and prevents identifying their pushforwards. □
Lemma A1
(Exceptional-fiber coefficient calculation). For the proper projection q : P 1 → pt and constant Tate coefficient Q P 1 H ( − 2 ) , interpreted in the usual unshifted cohomological convention,
H k q * Q P 1 H ( − 2 ) = Q H ( − 2 ) , k = 0 , Q H ( − 3 ) , k = 2 , 0 , otherwise .
If the input is placed in the perverse Hodge heart as Q P 1 H [ 1 ] ( − 2 ) , the output degrees are − 1 , 1 , respectively. These fiber groups do not compute the image of the corrected conifold connecting morphism.
Proof. 
The projective-line cohomology is H 0 ( P 1 ) = Q H and H 2 ( P 1 ) = Q H ( − 1 ) , with no other degrees. Tensor with the one-dimensional ( − 2 ) Tate object to obtain the first table. Translating the input by [ 1 ] translates output cohomology from k to k − 1 . □

Appendix C.3. One-Step Correspondence and Normalized Test Object

Let X − and X + be separated finite-type framed Hecke charts, Z μ the selected incidence correspondence, and
X − ← p Z μ → q X + .
Let K μ ∈ D b MHM ( Z μ ) include the chosen cohomological shift and Tate normalization. Define
H μ , * ( M ) = q * ( K μ ⊗ p * M ) , M ∈ D b MHM ( X − ) .
The operation q ! is different unless q is proper.
Lemma A2
(A normalized fiber test). Let q : P 1 → pt and use the pure Hodge module Q P 1 H [ 1 ] , of weight one, as a test kernel. Then
H − 1 q * Q P 1 H [ 1 ] ≃ Q H , H 1 q * Q P 1 H [ 1 ] ≃ Q H ( − 1 ) ,
with all other cohomology zero. This does not determine the normalization of the actual Hecke kernel K μ .
Proof. 
The underlying rational complex is R Γ ( P 1 , Q ) [ 1 ] . Its cohomology lies in degrees − 1 and 1, corresponding to H 0 ( P 1 , Q ) = Q and H 2 ( P 1 , Q ) = Q ( − 1 ) . The Hodge structures are pure of weights zero and two, respectively, giving (A43). The actual kernel may have an additional relative shift or Tate twist, which must be imported from its independent geometric definition rather than assigned by the fiber calculation. □

Appendix C.4. Two-Step Chains and Their Fibers

Let Λ 0 ⊃ Λ 1 ⊃ Λ 2 be a chain of two minuscule colength-one modifications. Let m : Gr ˜ μ , μ → Gr ¯ ( 2 , 0 ) retain the terminal lattice.
Proposition A18
(Two terminal strata). The module Q = Λ 0 / Λ 2 is either O / ( w 2 ) or ( O / ( w ) ) 2 . In the former case the intermediate lattice is unique; in the latter case the fiber of m is P 1 . The point with Q ≃ ( O / ( w ) ) 2 is the terminal lattice Λ 2 = w Λ 0 .
Proof. 
The structure theorem over the discrete valuation ring O gives the two length-two modules. Intermediate lattices correspond to length-one submodules of Q. The unique length-one submodule of O / ( w 2 ) is its socle w O / ( w 2 ) , while the length-one submodules of ( O / ( w ) ) 2 = C 2 are its one-dimensional vector subspaces. In the second case w Λ 0 ⊂ Λ 2 and comparison of colengths gives equality. This proves the claims. □
Proposition A19
(Higgs invariance depends on the intermediate field). Write the Higgs field in a chosen frame as A ( w ) = a b c d . After the first modification Λ 1 = w O e 1 ⊕ O e 2 , admissibility is b ∈ w O , and in the basis ( w e 1 , e 2 ) the induced field is
A 1 ( w ) = a b / w w c d .
For a second line [ u : v ] ∈ P ( Λ 1 / w Λ 1 ) , the invariance equation is
( w c ) u 2 + ( d − a ) u v − ( b / w ) v 2 ≡ 0 ( mod w ) .
The Higgs-compatible incidence fiber is the zero scheme of this binary quadratic, not necessarily the full P 1 fiber of Proposition A18.
Proof. 
The image of w e 1 belongs to Λ 1 automatically. The image of e 2 belongs to it precisely when b is divisible by w. Expressing the endomorphism in the new basis gives A 1 . The second line is invariant exactly when its representative ( u , v ) is linearly dependent with A 1 ( 0 ) ( u , v ) . Taking their determinant yields the displayed equation. □

Appendix C.5. The Second Higgs Line as a Scheme and a Nonflat Family

Choose the first lattice Λ 1 = w O e 1 ⊕ O e 2 . For a preserving matrix A ( w ) = a ( w ) w β ( w ) c ( w ) d ( w ) , its matrix in the adapted basis ( w e 1 , e 2 ) is A 1 ( w ) = a β w c d . Put a 0 = a ( 0 ) , d 0 = d ( 0 ) and β 0 = β ( 0 ) . A second line [ u : v ] ∈ P ( Λ 1 / w Λ 1 ) is invariant if and only if
v ( d 0 − a 0 ) u − β 0 v = 0 .
This is an equation of projective schemes, not just a condition on their closed points.
Proposition A20
(The three schemes and their two coefficient theories). The equation (A45) defines two distinct reduced points when d 0 ≠ a 0 , a doubled point Spec C [ t ] / ( t 2 ) when d 0 = a 0 and β 0 ≠ 0 , and the full P 1 when d 0 = a 0 and β 0 = 0 . In those three cases, respectively,
fiber { p , q } Spec C [ t ] / ( t 2 ) P 1 dim Q H 0 ( − , Q ) 2 1 1 dim C H 0 ( − , O ) 2 2 1
while H 2 ( P 1 , Q ) = Q ( − 1 ) is the only positive cohomology in the three cases. Thus a constructible Betti coefficient on the doubled point cannot by itself detect its nilpotent thickening, although the coherent structure sheaf does.
Proof. 
If d 0 − a 0 ≠ 0 , the two linear factors have different projective zeros. If d 0 = a 0 and β 0 ≠ 0 , the equation is a nonzero scalar times v 2 ; on u ≠ 0 , with t = v / u , it is t 2 = 0 . When both coefficients vanish, the equation is identically zero. A zero-dimensional finite scheme has constant rational cohomology in degree zero, with dimension the number of underlying geometric points; its coherent global sections are its coordinate algebra. For P 1 , use its standard coherent and rational cohomology. This proves the table. □
Proposition A21
(An explicit failure of flatness). For the family A τ ( w ) = 0 τ w 0 0 , the second-line scheme over A τ 1 has equation τ v 2 = 0 . Its generic fiber is the doubled point, while the special fiber at τ = 0 is P 1 . The family is not flat.
Proof. 
Here a 0 = d 0 = 0 and β 0 = τ . Equation (A45) is − τ v 2 = 0 . On the affine chart u ≠ 0 , put t = v / u . The coordinate ring is C [ τ , t ] / ( τ t 2 ) , in which the nonzero class t 2 is killed by τ . Hence this module over C [ τ ] has torsion and is not flat. Setting τ to a nonzero scalar or to zero gives the asserted fibers. □

Appendix C.6. The Full Punctual First-Jet Incidence and Terminal Image

Write the initial matrix as A ( w ) = w B + O ( w 2 ) , with B ∈ Mat 2 ( C ) . The first line ℓ ∈ P 1 is automatically preserved because A ( 0 ) = 0 . Put
β B ( ℓ ) = pr C 2 / ℓ ∘ B | ℓ ∈ Hom ( ℓ , C 2 / ℓ ) .
Globally β B is the section of Hom ( L , Q ) ≅ O P 1 ( 2 ) whose vanishing scheme is the eigenline scheme of B. For a second-line homogeneous coordinate [ u : v ] adapted to ℓ, the scheme of admissible chains has local equation
− β B ( ℓ ) v 2 = 0
inside the full chain surface F 2 . When B is nonscalar, the scheme contains the distinguished section with generic multiplicity two, as the factor v 2 shows; its reduction discards this thickening. If B is scalar, β B = 0 and the entire incidence is the reduced smooth surface F 2 . The equation transforms by the corresponding trivializations of the universal bundles; it does not define a global scalar function without these changes of frame.
Theorem A2
(Reduced punctual first-jet classification). Let Z B be the reduction of the scheme in (A47), let π : F 2 → P 1 be the first-line projection, and let S ⊂ F 2 be the section contracted by the terminal-lattice map. Then the reduced geometry is
  • Z B = S ∪ π − 1 ( ℓ 1 ) ∪ π − 1 ( ℓ 2 ) when B has two distinct eigenlines; its three rational curves form a tree;
  • Z B = S ∪ π − 1 ( ℓ 0 ) when B has a single eigenline and is nonscalar; its two rational curves meet in one point;
  • Z B = F 2 when B is scalar.
The rational mixed Hodge cohomology of these reduced fibers is, respectively,
H 0 H 2 H 4 two eigenlines Q H Q H ( − 1 ) 3 0 one nonscalar eigenline Q H Q H ( − 1 ) 2 0 scalar Q H Q H ( − 1 ) 2 Q H ( − 2 )
with all odd cohomology zero. This is the reduced punctual constant-coefficient incidence, not the full twice-modified Higgs target.
Proof. 
For fixed ℓ with β B ( ℓ ) ≠ 0 , reduction of (A47) forces v = 0 , giving the section S. When β B ( ℓ ) = 0 , the entire second-line projective fiber is allowed. The zero divisor of β B has two simple points, a double point, or is identically zero according as B has two distinct eigenlines, one nonscalar Jordan eigenline, or is scalar. Taking reduced schemes yields the three stated geometries. Distinct vertical fibers are disjoint, and each meets S at one point. Mayer–Vietoris for the resulting trees of projective lines gives one copy of Q H ( − 1 ) in H 2 for each irreducible component, and no odd cohomology. The cellular decomposition of F 2 yields the final row. □
Proposition A22
(Terminal-lattice direct image on the reduced incidence). Let m B : Z B → C B = m ( Z B ) be the terminal-lattice map, and write o B = m ( S ) . Its fibers are points off o B and the entire section S ≃ P 1 over o B . In the ordinary constructible cohomology-sheaf convention,
R 0 ( m B ) * Q Z B = Q C B , R 2 ( m B ) * Q Z B = Q { o B } ( − 1 ) ,
and the other higher direct images vanish. With coefficient Q ( − 2 ) , the displayed Tate types shift to Q ( − 2 ) and Q ( − 3 ) , respectively.
Proof. 
A terminal lattice whose quotient is O / ( w 2 ) has a unique intermediate lattice. The lattice w Λ 0 has a P 1 of intermediate lattices, which form S. The entire section is admissible when A ( 0 ) = 0 and is contracted by m. The remaining vertical components are mapped injectively off the section. Proper base change therefore identifies the stalk cohomology of the direct image with that of either a point or P 1 , giving the stated sheaves. This calculation contains no information about the transported extension morphism. □
Corollary A2
(Vanishing of the strictly punctual transported arrow). If the source map of a correspondence factors through the reduced node p, then its pullback of the corrected forward attaching arrow is zero by Proposition A5. All subsequent derived kernel and pushforward images of thatrestrictedarrow vanish. The punctual endpoint images (A49) consequently do not compute the numerical boundaries δ 02 k of the genuine two-step Higgs target, which retains transverse source directions.

Appendix C.7. The Extraordinary Costalk Test for the Genuine Two-Step Target

The reduced punctual terminal-lattice map m B and the genuine composed target q 02 are different morphisms. Let the latter be the finite-type algebraic correspondence of Definition 31, with Z 02 = Z 01 × X 1 Z 12 , p 02 : Z 02 → X 0 and q 02 : Z 02 → X 2 . For a closed target point y ∈ X 2 write F y = Z 02 × X 2 { y } , k y : F y → Z 02 , a y : F y → { y } and i y : { y } → X 2 . Set M R = K 02 ⊗ L p 02 * R − H and M I = K 02 ⊗ L p 02 * I − H .
Proposition A23
(Extraordinary proper-base-change formula). Suppose q 02 is proper on the selected coefficient support and the extraordinary base-change isomorphism holds in the chosen derived MHM setting. Then i y ! H 02 ( ϵ − ) is represented by
R a y * k y ! M R ⟶ R a y * k y ! M I [ 1 ] ,
where the arrow is R a y * k y ! ( 1 K 02 ⊗ p 02 * ϵ − ) . If this arrow is nonzero, the original composed arrow is nonzero. The formula does not assume an isomorphism k y ! ( K 02 ⊗ p 02 * M ) ≃ k y ! K 02 ⊗ p 02 , y * M .
Proof. 
The square formed by q 02 , i y , a y and k y is Cartesian. Extraordinary proper base change identifies i y ! R q 02 * with R a y * k y ! on the stated objects. Apply the exchange isomorphism naturally to the two endpoints and the arrow between them. Every functor sends zero morphisms to zero, so nonvanishing of (A50) forces nonvanishing of the original arrow. □
Theorem A3
(Smooth-purity criterion for two-step nonvanishing). In the setting of Proposition A23, assume that ϵ − = e * ϵ o for an evaluation e : X 0 → X det , and put t = e ∘ p 02 : Z 02 → X det . Suppose that q 02 is proper on the displayed coefficients, that F y is nonempty, that t is smooth of constant relative dimension r on a neighborhood of F y , and that t ∘ k y factors through the node i o : { o } ↪ X det as t ∘ k y = i o ∘ a y . Assume, moreover, that on a neighborhood of F y the composed coefficient kernel is the shifted and twisted tensor unit K 02 ≃ Q Z 02 H [ d ] ( m ) , with this isomorphism specified on the actual arrow. If H 0 ( i o ! ϵ o ) = c o id with c o ≠ 0 , then
i y ! H 02 ( ϵ − ) ≠ 0 , H 02 ( ϵ − ) ≠ 0 .
Neither the assumption F y = t − 1 ( o ) nor smoothness of the fiber F y is required. The conclusion concerns the full derived attachment arrow, not any specified perverse boundary δ 02 k .
Proof. 
The unit-kernel and shift/twist assumptions reduce the morphism inside the extraordinary fiber formula to k y ! t * ϵ o [ d ] ( m ) . For a smooth morphism of relative dimension r, purity gives t ! ≃ t * ( r ) [ 2 r ] , or equivalently t * ≃ t ! ( − r ) [ − 2 r ] . The factorization t k y = i o a y and pseudofunctoriality of extraordinary pullback therefore give a natural comparison of the specified arrow
k y ! t * ϵ o [ d ] ( m ) ≃ k y ! t ! ϵ o [ d − 2 r ] ( m − r ) ≃ a y ! i o ! ϵ o [ d − 2 r ] ( m − r ) .
In particular this exchange does not commute k y ! past a general tensor product: it works here only because the specified kernel is the shifted and twisted tensor unit. By the projection formula for a y and its dualizing object ω F y = a y ! Q H ,
R a y * a y ! i o ! ϵ o ≃ i o ! ϵ o ⊗ L R a y * ω F y .
Since F y is nonempty and proper, duality for its structure map gives R a y * ω F y ≃ D pt R Γ ( F y , Q F y H ) , whose degree-zero cohomology is the nonzero rational vector space dual to H 0 ( F y , Q ) . The underlying rational realization of H 0 ( i o ! ϵ o ) is the nonzero scalar map c o id on a one-dimensional vector space. Bounded complexes of rational vector spaces split in the derived category into their cohomology spaces, so the derived tensor of this morphism with the nonzero complex R a y * ω F y is nonzero: on the tensor summand formed from the source degree-zero costalk and the degree-zero part of the dualizing cohomology it induces c o id ⊗ id ≠ 0 . Faithfulness of rational realization on morphisms detects this nonzero image in derived mixed Hodge structures. Apply Proposition A23 and restore the common shift [ d − 2 r ] ( m − r ) to conclude (A51). □
Remark A14
(Geometric verification and numerical boundaries). The theorem gives checkable sufficient hypotheses, rather than assuming nonvanishing after the fiber pushforward. It is not an unconditional assertion about the actual iterated Higgs charts: the composite evaluation t = e ∘ p 02 must be smooth near the chosen target fiber, and the actual composed kernel must have the specified shifted constant form. The punctual incidence trees in Theorem A2 are candidate fibers; their Betti cohomology does not prove either of these hypotheses. The formula t ∘ k y = i o a y suffices for the extraordinary pullback calculation when smooth purity applies; equality of the target fiber with the full inverse image of the node is not needed. For a nonconstant kernel a separate k y ! -tensor exchange and nonvanishing calculation remain necessary. Finally, the theorem detects a nonzero derived arrow but does not identify its numerical perverse cohomology maps δ 02 k or its Satake-channel projections.

Appendix C.8. Convolution Kernels and the Composition Square

Let X 0 ← p 01 Z 01 → q 01 X 1 and X 1 ← p 12 Z 12 → q 12 X 2 be the finite-type correspondences of Definition 29. Let Z 02 = Z 01 × X 1 Z 12 with projections a : Z 02 → Z 01 , b : Z 02 → Z 12 . Write K 02 = a * K 01 ⊗ b * K 12 .
Proposition A24
(The four functorial steps of convolution). Assume proper Beck–Chevalley for the Cartesian square and the projection formula for the selected kernels. Then the natural equivalence of the two-step and composed operators is the composite of the following named maps:
  • the base-change isomorphism p 12 * q 01 * → ∼ b * a * ;
  • the projection-formula isomorphism K 12 ⊗ b * ( − ) → ∼ b * ( b * K 12 ⊗ − ) ;
  • pseudofunctoriality of pullback and pushforward, a * p 01 * ≃ ( p 01 a ) * and q 12 * b * ≃ ( q 12 b ) * ;
  • the symmetry/associativity isomorphism of derived tensor products identifying the resulting kernel with K 02 .
For every source arrow ϵ − : R − → I − [ 1 ] these maps assemble into the naturality square
H 12 H 01 ( R − ) → H 12 H 01 ( ϵ − ) H 12 H 01 ( I − ) [ 1 ] ∼ κ R − H ↓ ∼ κ R − H κ I − H [ 1 ] ↓ κ I − H [ 1 ] H 02 ( R − ) → H 02 ( ϵ − ) H 02 ( I − ) [ 1 ] .
Proof. 
Expand the composite on an object M:
q 12 * K 12 ⊗ p 12 * q 01 * ( K 01 ⊗ p 01 * M ) .
Apply base change to replace p 12 * q 01 * with b * a * . Apply the projection formula to move K 12 inside b * . Apply pseudofunctoriality and the symmetry of tensor product. The result is
( q 12 b ) * ( a * K 01 ⊗ b * K 12 ) ⊗ ( p 01 a ) * M = H 02 ( M ) .
Each named transformation is natural in M, hence their composite is a natural equivalence of exact functors. Evaluate that natural equivalence on ϵ − ; exactness identifies the component at I − [ 1 ] with the shift of the component at I − . This is exactly (A54). Its validity does not assert nonvanishing of either horizontal arrow. □

Appendix C.9. Chain Maps, Homotopy, and the Explicit Octahedron

A composition comparison on an attaching arrow must use a consistent cone convention. Let f : A • → B • and g : B • → C • be cochain maps, with Cone ( f ) n = B n ⊕ A n + 1 and differential d B f 0 − d A . Put F = Cone ( f ) , H = Cone ( g f ) and G = Cone ( g ) .
Proposition A25
(An explicit octahedral quasi-isomorphism). The degree-zero maps
u : F → H , ( b , a ) ↦ ( g b , a ) , v : H → G , ( c , a ) ↦ ( c , f a )
are cochain maps. Their composite has a degree-minus-one null-homotopy
h : F n → G n − 1 , h ( b , a ) = ( 0 , b ) , d G h + h d F = v u .
The map
w : Cone ( u ) ⟶ G , w ( c , a , b ′ , a ′ ) = ( c , f a + b ′ )
(with c ∈ C n , a ∈ A n + 1 , b ′ ∈ B n + 1 , a ′ ∈ A n + 2 ) is a cochain quasi-isomorphism. Consequently the triangle
F → u H → v G ⟶ F [ 1 ]
is distinguished in the derived category.
Proof. 
Substitution into the cone differentials gives d H u = u d F and d G v = v d H . On ( b , a ) ∈ F n the composite is v u ( b , a ) = ( g b , f a ) . A direct calculation gives
d G h ( b , a ) = ( g b , − d B b ) , h d F ( b , a ) = ( 0 , d B b + f a ) ,
whose sum is ( g b , f a ) . The mapping cone of u has the four components shown in (A57), with differential
d ( c , a , b ′ , a ′ ) = ( d C c + g f ( a ) + g ( b ′ ) , − d A a + a ′ , − d B b ′ − f ( a ′ ) , d A a ′ ) .
Substitution into w shows w d = d G w , so w is a chain map. It is surjective in every degree: ( c , b ′ ) ∈ G n is the image of ( c , 0 , b ′ , 0 ) . Its kernel consists of the elements ( 0 , a , − f a , a ′ ) . In these parameters its differential is
D ( a , a ′ ) = ( − d A a + a ′ , d A a ′ ) , H ( a , a ′ ) = ( 0 , a ) , D H + H D = id .
Hence ker w is contractible. The short exact sequence 0 → ker w → Cone ( u ) → w G → 0 is a quasi-isomorphism, and the cone triangle of u becomes (A58); the second map is v because w restricts to v on H. □
Remark A15
(Relationship to the actual Hecke composition). This is a chain-level statement in a category where the morphisms have compatible cochain representatives. Applying it to a particular Hecke coefficient diagram requires constructing the actual composable maps f , g and the identifications of their cones with the geometric defect complexes. The abstract quasi-isomorphism does not compute the surviving image of ϵ − under two-step convolution.

Appendix C.10. The Satake Tensor-Square and the Genuine Comparison Boundary

Let G = G L 2 and let V be the standard representation of G ∨ ≃ G L 2 . In Rep Q ( G ∨ ) ,
V ⊗ V ≃ Sym 2 V ⊕ det V .
The summands have dimensions three and one. Let S μ be the normalized minuscule spherical Hodge object under Fedorov’s pure Hodge–Satake equivalence.
Proposition A26
(The decomposition inside the spherical category). The convolution S μ ★ S μ decomposes into the two Satake objects corresponding to Sym 2 V and det V , with the parity/Tate normalization fixed by the Hodge–Satake equivalence [11,25]. This is an isomorphism in the spherical Hodge category; it does not supply endomorphisms of the unrelated Higgs-modification kernel.
Proof. 
The Hodge–Satake tensor equivalence transports convolution to tensor product of dual-group representations, with its prescribed Tate and parity convention. Apply it to the standard representation and the G L 2 decomposition (A61). Since the equivalence is additive, it transports the direct sum to the stated convolution decomposition. The functor is defined on the spherical category only; therefore the conclusion does not apply to a Higgs kernel without an independent comparison. □
Remark A16
(The two projected attachment arrows remain conditional). Only after a geometric and coefficient-level Satake-to-Higgs comparison β μ has been constructed, and only after its convolution square carries the two idempotents to the actual composed Higgs kernel, may one define ϵ sym and ϵ det as in (165). Under those conditions the two components satisfy the formal direct-sum identity of Proposition 53. Neither component’s vanishing, nor its stalk/costalk degree, follows from (A61). No numerical channelwise attachment calculation is claimed without the kernel comparison and the induced maps on the specified extension.

Appendix D. Stabilized A 2 Specialization and Stokes Data

The stabilized A 2 family of Section 2.8 supplies a finite and explicit spectral collision. The critical-point and critical-value calculations are algebraic. A Stokes matrix requires ordered distinguished paths and oriented thimbles, while any comparison with the corrected mixed-Hodge-module attachment requires an independently defined functor. We keep these assertions separate.

Appendix D.1. Critical Locus, Values, and Jacobian Algebra

Let
F s ( x , y , z , w ) = x 3 3 − s x + y 2 + z 2 + w 2 .
The relative Jacobian algebra over C [ s ] is
J F = C [ s , x , y , z , w ] ( x 2 − s , 2 y , 2 z , 2 w ) ≃ C [ s , x ] / ( x 2 − s ) .
Proposition A27
(Critical-value cover and spectral operator). The critical points over s ≠ 0 are q ± = ( ± s , 0 , 0 , 0 ) , with critical values T ± = ∓ 2 3 s 3 / 2 . The critical-value curve is
T 2 = 4 9 s 3 .
Multiplication by the critical-value function in the free C [ s ] -module J F , in basis ( 1 , x ) , is represented by
D s = 0 − 2 3 s 2 − 2 3 s 0 , D s 2 = 4 9 s 3 id .
Proof. 
The partial derivatives are ∂ x F s = x 2 − s and ∂ y F s = 2 y , ∂ z F s = 2 z , ∂ w F s = 2 w , yielding (A63). At a critical point s = x 2 and F s = − 2 3 x 3 ; substituting x = ± s gives the two values and (A64). In the Jacobian algebra x 3 = s x , so multiplication by F s equals multiplication by − 2 3 s x . It sends 1 to − 2 3 s x and x to − 2 3 s 2 , giving the matrix in (A65). Squaring it yields 4 9 s 3 id . □

Appendix D.2. Projectors, Normalization, and Descent

Pass to the punctured base and choose a branch r = s . Let K ′ = C ( ( r ) ) and R ′ = C [ [ r ] ] . The eigenvalues of D s are λ ± = ∓ 2 3 r 3 .
Proposition A28
(Generic eigenprojectors and the descent obstruction). On the generic two-dimensional vector space over K ′ , the spectral idempotents are
e + = D s − λ − λ + − λ − , e − = D s − λ + λ − − λ + .
They satisfy e ± 2 = e ± , e + e − = 0 , and e + + e − = 1 . A loop about s = 0 exchanges their labels. The denominators in (A66) do not alone prove that the projectors fail to extend to every normalized lattice.
Proof. 
The characteristic polynomial is T 2 − 4 9 s 3 , with distinct eigenvalues on s ≠ 0 . Lagrange interpolation gives the stated expressions, which are orthogonal idempotents by direct substitution using the characteristic polynomial. Analytic continuation changes r ↦ − r and interchanges the two eigenvalues, hence their labeled projectors. In an eigenbasis over K ′ the operators are constant diagonal idempotents, so they extend on the diagonal lattice R ′ 2 . Such a lattice need not descend to the original s-disc. Thus labeled descent and lattice extension are separate issues. □

Appendix D.3. An Integral-Lattice Test for the Generic Spectral Projectors

The denominators of the interpolation projectors can be tested against a specified lattice instead of being treated only as an informal warning. Use the natural Jacobian lattice J = C [ [ s ] ] 〈 1 , x 〉 with multiplication operator D s as in (A65). After the ramified base change s = r 2 , set J ′ = C [ [ r ] ] 〈 1 , x 〉 .
Proposition A29
(Failure on the natural Jacobian lattice). Over C ( ( r ) ) the normalized operator
Q = D s ( 2 / 3 ) r 3 = 0 − r − r − 1 0 , Q 2 = 1 ,
has idempotents e ± = ( 1 ± Q ) / 2 . Neither preserves the specified lattice J ′ , because their lower-left entries contain ∓ ( 2 r ) − 1 . This is a genuine nonextension resultfor that lattice; it does not prohibit extension on a different lattice spanned by eigenvectors.
Proof. 
Substitute s = r 2 into D s = 0 − 2 3 s 2 − 2 3 s 0 and divide by ( 2 / 3 ) r 3 to obtain (A67). Matrix multiplication gives Q 2 = 1 , so e ± 2 = e ± , e + e − = 0 , and e + + e − = 1 . Since e ± ( 1 ) contains a nonintegral r − 1 x term, e ± ( J ′ ) ¬ ⊆ J ′ . Over the fraction field, both projectors are diagonal in an eigenbasis; the span of those eigenvectors gives a different lattice on which they are integral. A labeled projector on the s-disc requires additional Z / 2 descent data because r ↦ − r exchanges the eigenvalues. □

Appendix D.4. Distinguished Paths and Stokes/Picard–Lefschetz Data

Fix s ≠ 0 and a regular reference value T 0 outside the two critical values. Choose two nonintersecting distinguished paths γ + , γ − from T 0 to T + , T − with a fixed ordering. Let Δ ± be their oriented Lefschetz thimbles and δ ± the associated vanishing cycles in the reference fiber. Since F s has complex dimension four as a function, δ ± are real three-dimensional vanishing spheres.
Definition A2
(Ordered Picard–Lefschetz/Stokes datum). The datum consists of the ordered paths, orientations of Δ ± , the intersection pairing 〈 δ + , δ − 〉 , and the choice of sectors in the Fourier–Laplace plane in which the exponentials exp ( T + / ℏ ) and exp ( T − / ℏ ) are ordered. A Stokes wall occurs where Re ( ( T + − T − ) / ℏ ) = 0 .
Proposition A30
(Dependence of the elementary Stokes factor). For an ordered distinguished basis the elementary Stokes change is unipotent triangular. Its off-diagonal coefficient is the signed Picard–Lefschetz intersection number determined by the paths and orientations. Reversing the orientation of one vanishing cycle changes the sign of this coefficient. The unordered critical-value curve (A64) does not determine a labeled Stokes matrix.
Proof. 
Picard–Lefschetz transport across a distinguished vanishing path acts by a transvection of the form
v ⟼ v + σ 〈 v , δ 〉 δ ,
where the fixed sign σ is determined by the dimension and the chosen intersection and orientation convention. Expressing this transformation in the ordered basis gives a triangular matrix with diagonal entries one and the indicated signed intersection coefficient. Replacing δ by − δ changes the labeled basis and its off-diagonal entry by sign; reordering the paths also changes which triangular factor is used. The equation of the critical-value curve records only the unordered branch locus and therefore does not encode these choices. □
Remark A17
(Comparison with a Hecke attachment is not automatic). The Jacobian algebra is a finite algebraic model for critical values; its multiplication operator is not a morphism in D b MHM ( X 0 ) without an independently constructed comparison. The irregular Stokes category and the ordinary regular-holonomic MHM category also differ. In particular, neither branch exchange nor a nontrivial Stokes factor proves that a specific conifold Yoneda extension survives Hecke convolution.

Appendix D.5. A Precise Rank-Two Stokes Test and Its Limitations

We can calculate the adjacency of the two thimbles before introducing the three quadratic suspensions. Take real s > 0 and the regular reference value T 0 = 0 . The cubic x 3 / 3 − s x = 0 has three distinct real roots
r 1 = − 3 s , r 2 = 0 , r 3 = 3 s .
Choose paths to the two real critical values so that their vanishing zero-cycles in the reduced H 0 of the three-point fiber are δ 1 ( 0 ) = [ r 1 ] − [ r 2 ] and δ 2 ( 0 ) = [ r 2 ] − [ r 3 ] . With the standard dot pairing on the free group generated by the three points, the adjacent pairing is computed directly:
δ 1 ( 0 ) , δ 2 ( 0 ) = 〈 e 1 − e 2 , e 2 − e 3 〉 = − 1 .
Stabilization by y 2 + z 2 + w 2 suspends the two vanishing cycles to real three-spheres. Under the standard Picard–Lefschetz suspension rule their middle-dimensional intersection pairing has the same absolute value, namely one; its sign depends on the chosen orientation and intersection-form conventions. Consequently the elementary ordered unipotent factor has the form
S γ , δ = 1 σ 0 1 , σ ∈ { + 1 , − 1 } ,
up to the adopted upper/lower-triangular ordering. This displayed matrix is a normalized Picard–Lefschetz test, not an intrinsically labeled Stokes matrix of the unordered cusp.
Remark A18
(What is required for an irregular comparison). A claim that (A68) is the Stokes matrix of a specific Fourier–Laplace module additionally requires the actual irregular connection, a sector ordering, comparison with its distinguished thimble basis and its orientation conventions. It also requires a separate functor to compare that irregular object with a Saito regular-holonomic conifold attachment. The finite Jacobian and projector calculations above supply neither such a functor nor the image of ϵ Σ .

Appendix E. Magnetic Hecke Admissibility and Coefficient Naturality

The extended-Bogomolny scattering correspondence supplies a geometric interface with Hecke modifications under its analytic hypotheses. The calculations below distinguish that interface from a functor on derived mixed Hodge modules and specify the actual morphism-level data required to transport an attachment. The automorphic Betti spectral-action comparison is outside the scope of this appendix, and no nilpotent singular-support or automorphic graph-kernel construction is assumed.
Remark A19
(Star and proper-support transport). For a specified algebraic Hecke correspondence X − ← p Z → q X + and fixed kernel K , the two operators q * ( K ⊗ p * ( − ) ) and q ! ( K ⊗ p * ( − ) ) need not agree when q is nonproper. The pullback p ! is a further distinct normalization. A geometric scattering correspondence by itself supplies none of these coefficient functors on a gauge-theoretic category; the source and target, the chosen operation, and its extension arrow must be defined independently.

Appendix E.1. Magnetic Hecke Scattering: Analytic Input

Let M = ( 0 , 1 ) × C with a marked point ( t 0 , x ) , let G be compact with complexification G C , and let μ be a fixed dominant coweight prescribing the singularity type of a magnetic modification. Write M Bog ( M , μ ) for a gauge-theoretic moduli space only after specifying its boundary conditions, gauge group, regularity class, and stability assumptions.
Datum Required check
Base and puncture ( 0 , 1 ) × C and the specified singular point ( t 0 , x )
Coweight Dominant μ and its allowed Hecke relative position
Boundary conditions Exact Nahm-pole/other boundary conditions and behavior at t = 0 , 1
Gauge quotient Gauge group, framing, stabilizers, and regularity of the quotient
Analytic hypotheses Irreducibility/stability, solution class, and compactness assumptions of the chosen theorem
Scattering map Well-defined map into the Hecke-modification moduli, with its exact range and equivalence status
Coefficient action Separately constructed derived functor on the specified attachment-bearing category
Remark A20
(Scope of the scattering comparison). The correspondence between the extended Bogomolny equations and Hecke modifications is invoked only within the analytic hypotheses of [14]. A scattering correspondence on solutions or moduli points does not define a functor on mixed Hodge modules; that requires suitable algebraic geometry and coefficient kernels. The type-IIB conifold N = 2 light-sector description is not identified with twisted N = 4 magnetic line operators [15,26].

Appendix E.2. The Extended Bogomolny Equations Underlying the Scattering Map

Fix the product three-manifold W = [ 0 , 1 ] × C with one interior singular point ( y 0 , x ) , a compact structure group G, a principal bundle on the punctured manifold, and a unitary connection A. In the convention used in Section 8.3, let ϕ be an adjoint-valued one-form and ξ an adjoint-valued zero-form. The extended Bogomolny system is
F A − 1 2 [ ϕ ∧ ϕ ] = * d A ξ , d A ϕ − * [ ξ , ϕ ] = 0 , d A * ϕ = 0 .
The scattering map of [14] is applied only to the specific boundary, Dirac singularity, stability, and gauge-group conditions under which its correspondence with Hecke modifications is established. For a lower minuscule modification the associated line must satisfy A ( 0 ) ℓ ⊆ ℓ on the Higgs side, so its closed-point image lies in the scheme of (215). The analytic identification on solution classes supplies no functor from a gauge-theoretic coefficient category to D b MHM ( X ) , nor a comparison on the nonreduced infinitesimal incidence structure; either assertion requires its own functor on morphisms and comparison maps.

Appendix E.3. A Conditional Line-Operator Naturality Square

Definition A3
(Required coefficient-level magnetic comparison). Suppose C − and C + are independently defined stable coefficient categories, T μ : C − → C + is a magnetic operation, and R ± : C ± → D ± are exact functors to the selected derived MHM categories. A magnetic comparison is a specified natural transformation
Θ μ mag : R + T μ ⟹ H μ H R − ,
where H μ H : D − → D + is the independently defined Hecke pull–push operator. For an arrow α : A 1 → A 0 [ 1 ] , naturality requires
R + T μ ( A 1 ) → R + T μ ( α ) R + T μ ( A 0 ) [ 1 ] Θ μ , A 1 mag ↓ Θ μ , A 1 mag Θ μ , A 0 mag [ 1 ] ↓ Θ μ , A 0 mag [ 1 ] H μ H R − ( A 1 ) → H μ H R − ( α ) H μ H R − ( A 0 ) [ 1 ] .
Proposition A31
(Soundness on the selected arrow). If Θ μ mag is a defined natural transformation of exact functors, then (A70) commutes in the homotopy category and admits the corresponding homotopy-coherent square in the enhancements. This statement does not assert that the functors or transformation exist for the Bogomolny scattering construction.
Proof. 
Apply naturality of Θ μ mag to α . Exactness identifies its component at A 0 [ 1 ] with the shift of its component at A 0 , yielding the square. The coherent transformation in the enhanced functor category supplies its homotopy. No geometric existence claim is used. □
Remark A21
(Additional duality data). A Wilson/’t Hooft comparison requires a separately constructed duality functor S and a Wilson operator W V , together with a natural transformation comparing S ∘ T μ with W V ∘ S and its action on α. Neither the bounded Hecke correspondence nor the local conifold MHM extension furnishes these functors. No S-duality or line-operator equivalence is asserted here.

Appendix E.4. Final Dependency Ledger

Remark A22
(Use of the appendices in subsequent work). The formal F-bundle proof package in Appendix F is separate from the conifold Rees specialization. The local intersection-complex costalk and the one-dimensional Ext group in Appendix A, the strict Rees sequence and the specialization map in Appendix B, and the finite Hecke-lattice and base-change calculations in Appendix C supply the controlled MHM and filtered inputs. A two-step geometric correspondence and its natural comparison act on the specified conifold arrow under the hypotheses of Section 6. The genuine second-target numerical boundary requires its actual target and coefficient morphism. The extraordinary fiber formula and the smooth-purity nonvanishing criterion are available in Propositions A23 and A3, respectively; their application to the full Higgs target still requires verification of the actual smoothness and kernel conditions. By contrast, local nonvanishing after Rees specialization for the determinantal ODP is now proved in Theorem 3; only the exact marked R2 scalar remains unspecified.
The spherical Satake-to-Higgs kernel map, an independent matrix-factorization/Stokes-to-MHM functor, a motivic or Nori lift, and a magnetic coefficient functor are additional construction problems. None is assumed or needed for the internal MHM composition identity.

Appendix F. Formal F-Bundle Rees Constructions and the Categorical Comparison Boundary

This appendix supplies the local calculations supporting Theorem 6. The algebraic Rees parameter z of Appendix B is not identified with the auxiliary connection parameter u used here. The former homogenizes a filtered differential module on a smooth ambient variety; the latter occurs in the meromorphic connection defining an F-bundle over a base. The categorical source, target, and maps must be retained before composing either construction with a proposed singular-to-smooth comparison.

Appendix F.1. A Local Adapted-Basis Construction and Pole Orders

Let B be a smooth base over a characteristic-zero field k, and let ( E , ∇ B , F • E ) be an object of FilFlat k GT ( B ) as in Definition 36. Work over an open V ⊂ B on which the decreasing filtration is split by a basis e 1 , … , e r . Write p a for the last index with e a ∈ F p a E and set
g a = u − p a e a , H = ⨁ a = 1 r O V [ [ u ] ] g a .
Any section of F p E is spanned by those e a with p a ≥ p , so u − p e a = u p a − p g a has nonnegative power of u. Conversely each g a occurs in u − p a F p a E . This proves the lattice equality (179) and its local freeness without an assumed embedding into a quantum cohomology bundle.
Write ∇ B , ξ e a = ∑ b Γ b a ( ξ ) e b . Griffiths transversality implies Γ b a = 0 when p b < p a − 1 . The actual connection matrix on (A71) is
∇ ˜ ξ g a = ∑ b Γ b a ( ξ ) u p b − p a g b , ∇ ˜ ∂ u g a = − p a u g a .
The exponents in the base directions are at least − 1 . The auxiliary-direction connection has only a logarithmic pole, not an order-two term. Thus u ∇ ˜ ξ and u 2 ∇ ˜ ∂ u preserve the lattice.
Proposition A32
(The two special-fiber operators). Using the special-fiber identification
H / u H = ⨁ p gr F p E ,
we obtain
μ ξ ( [ s ] p ) = [ ∇ B , ξ s ] p − 1 , κ = 0 .
The first map is well-defined because changing s by t ∈ F p + 1 E changes ∇ B , ξ s by an element of F p E , and the second follows from u 2 ∇ ˜ ∂ u ( g a ) = − p a u g a .
Proof. 
The base-direction formula (A72), multiplied by u and reduced modulo u, retains only the terms for which p b = p a − 1 . These are precisely the components of ∇ B , ξ on F p E / F p + 1 E → F p − 1 E / F p E . The operator u 2 ∇ ˜ ∂ u sends every generator g a into u H ; by the Leibniz rule it sends an arbitrary u-adic coefficient multiple of g a there as well. Hence its reduction modulo u is zero. □

Appendix F.2. The Exact Functor on Morphisms and Extension Arrows

A horizontal filtration-preserving map h : ( E , ∇ , F ) → ( E ′ , ∇ ′ , F ′ ) acts on the punctured product by h ( ( u − p s ) ) = u − p h ( s ) . Since h ( F p E ) ⊆ F ′ p E ′ , the formula preserves Rees lattices; since h is horizontal and independent of u, it commutes with both connection directions. Direct substitution verifies R u ( id ) = id and R u ( k h ) = R u ( k ) R u ( h ) . This proves functoriality on the actual morphisms, not merely on isomorphism classes.
Proposition A33
(Strict sequences and the boundary map). Let 0 → E A → a E P → b E C → 0 be a strict short exact sequence of filtered flat bundles, with the filtration-level sequences
0 ⟶ F p E A ⟶ F p E P ⟶ F p E C ⟶ 0 ( p ∈ Z )
exact and with locally free quotients. Then
0 ⟶ R u ( E A ) → R u ( a ) R u ( E P ) → R u ( b ) R u ( E C ) ⟶ 0
is a strict sequence of formal F-bundles. The induced exact derived functor carries the connecting arrow ϵ F : E C → E A [ 1 ] to the boundary of (A75); it does not, without an additional detection theorem, preserve its nonvanishing.
Proof. 
On each open where the filtered short exact sequence is strictly split as a sequence of filtered bundles, lift an adapted basis of E C through every filtration level in (A74) and adjoin an adapted basis of E A . This gives a basis of E P compatible with its filtration and exhibits the Rees sequence as split over the underlying O B [ [ u ] ] -modules. Such local splittings need not commute with the connection; horizontal splitting is not asserted. Therefore the displayed sequence is strict exact in the quasi-abelian formal F-bundle category. The derived functor associated with an exact functor of exact categories sends a conflation to a distinguished triangle. Naturality of the connecting morphism gives the boundary assertion. A nonfaithful map on degree-one derived Hom groups may kill the boundary even though the original extension was nonsplit. □

Appendix F.3. A Two-by-Two Calculation and Exponential Shifts

Take B = Spec k [ t ] , a trivial rank-two bundle E = O B e 1 ⊕ O B e 2 , the connection ∇ B , ∂ t e 1 = e 2 , ∇ B , ∂ t e 2 = 0 , and the filtration F 1 E = O B e 1 , F 0 E = E , F 2 E = 0 . The connection is flat in this one-dimensional base, and ∇ ( F 1 E ) ⊆ F 0 E ⊗ Ω B 1 . The Rees generators are g 1 = u − 1 e 1 and g 2 = e 2 , and the horizontal operators satisfy
∇ ˜ ∂ t g 1 = u − 1 g 2 , ∇ ˜ ∂ t g 2 = 0 , ∇ ˜ u 2 ∂ u g 1 = − u g 1 , ∇ ˜ u 2 ∂ u g 2 = 0 .
Consequently, in the ordered basis of the fiber at u = 0 ,
μ ∂ t = 0 0 1 0 , κ = 0 0 0 0 .
The first operator is nonzero; the Euler spectral operator is zero. This example also fails the maximality condition of KKPY because the base dimension is one and the bundle rank is two.
For the rank-one exponential twist whose connection is ∂ u + c / u 2 , tensor-product connections give
κ H ⊗ L c = κ H + c id .
Thus a single scalar exponential twist of this elementary Rees bundle still has scalar Euler operator. A quantum F-bundle with nonscalar Euler operator cannot be isomorphic to such an object through a horizontal morphism regular and invertible at u = 0 , because any such morphism must intertwine the two Euler operators on the special fiber. This does not exclude more general constructions or distinct twists of independently constructed spectral summands.
Remark A23
(Calabi–Yau Euler multiplication and the scope of the obstruction). At a quantum point where the Euler multiplication Eu ★ − is zero, the specific obstruction comparing κ = 0 with a nonzero Euler operator does not apply. In the Calabi–Yau setting this can occur at a designated small-quantum point, depending on the grading and parameter conventions of the quantum connection. Vanishing of this single operator neither establishes equality of the complete meromorphic connections nor constructs an extension-preserving functor. A comparison with the conifold-transition framework of Lee–Lin–Wang also requires an actual map between the geometric/Hodge constructions and the independently selected quantum coefficient objects: cohomological transition identities do not supply the marked square (A83).

Appendix F.4. Selected Spectral Idempotents on Strict Extensions

Let 0 → A → a P → b C → 0 be a strict exact sequence in a Karoubi-closed category of F-bundles. Suppose horizontal idempotents e A , e P , e C satisfy e P a = a e A and b e P = e C b . Splitting the idempotents block-diagonalizes the two maps and gives the two sequences
0 → A e → P e → C e → 0 , 0 → A 1 − e → P 1 − e → C 1 − e → 0 .
Each is strict exact because the original conflation is their direct sum. On their connecting maps, naturality yields
e A [ 1 ] ϵ = ϵ e C .
An idempotent on the special fiber H / u H is not sufficient: the required endomorphism must lift to a horizontal idempotent of the entire F-bundle in the specified category, and it must respect both extension maps. KKPY atomicity additionally requires their maximality, equivariance and single-Euler-eigenvalue conditions. A spectral marking is therefore not itself a KKPY global atom, and the full spectrally marked subcategory is not asserted to be extension-closed.

Appendix F.5. The Missing Source-to-Target Morphism and Its Precise Test

For the fixed smooth ambient embedding i : X 0 ↪ Y , Saito provides the exact filtered-component functor and the algebraic Rees operation
D b MHM ( X 0 ) ⟶ D b ( FilRH str ( Y ) ) ⟶ D gr b ( R F D Y ) .
The constructed filtered-flat Rees functor has different source and target:
D ex b FilFlat k GT ( B ) → D b R u D ex b Fbun k for ( B ) .
These arrows do not compose. The first route has a potentially singular regular-holonomic D Y -module; the second requires a locally free flat connection over a specified base B. Neither route alone identifies an analytic quantum F-bundle over a non-archimedean base B F .
Remark A24
(Possible broader integrable-Rees target). The support obstruction for the conifold localization module rules out an identification with an ambient nonsingular flat vector bundle, not with every category of integrable Rees modules having singular support. The latter would have to be fixed as a separate exact or derived coefficient category and supplied with a compatible auxiliary-direction connection and a map to the proposed F-bundle category. Simply setting z = u does not give such a functor. This distinction does not alter the proved non-splitting of the conifold associated-graded extension.
Proposition A34
(Smooth-locus annihilation of the punctual arrow). Let j : U = X 0 ∖ Σ ↪ X 0 and let ϵ Σ : R Σ H → I H [ 1 ] be the corrected attachment. For every exact functor T : D b MHM ( U ) → E into a stable category,
( T ∘ j * ) ( ϵ Σ ) = 0 .
Proof. 
The quotient R Σ H is supported on Σ , so j * R Σ H = 0 . The source of j * ϵ Σ is the zero object, hence j * ϵ Σ = 0 . Every functor sends a zero arrow to a zero arrow, proving the stated formula. The argument applies to restriction to the smooth locus of the central fiber; a functor constructed from nearby smooth fibers and nearby/vanishing cycles need not factor through this restriction and is not excluded by the proof. □
To obtain the proposed singular-to-smooth realization one must construct an exact functor on a geometrically selected domain containing I H , P H , R Σ H , define it on all required morphisms, and independently specify a strict extension in the appropriate KKPY atomic F-bundle category. With fixed endpoint comparisons η I , η R , the required test is the actual commuting square
η I [ 1 ] R at ( ϵ Σ ) = ϵ Q η R .
Before R at and ϵ Q have been independently constructed, (A83) is a test specification, not an established identity. In particular, neither the successful filtered-flat Rees construction nor the matching dimension of two spectral objects establishes it. The relevant degeneration data, monodromy and adjunction morphisms must enter any proposed object and morphism assignment if it is to detect the punctual conifold attachment.

References

  1. Katzarkov, L.; Kontsevich, M.; Pantev, T.; Yu, T.Y. Birational Invariants from Hodge Structures and Quantum Multiplication, 2026. arXiv 2026, arXiv:math. [Google Scholar]
  2. Rahman, A. Mixed Hodge Modules and Canonical Perverse Extensions for Multi-Node Conifold Degenerations, 2026. arXiv 2026, arXiv:math. [Google Scholar]
  3. Rahman, A. Bivariant Hodge Atoms I: Hecke Collisions and Mixed-Hodge Coefficient Extensions, 2026. Preprints.org 2026, version 1. [Google Scholar] [CrossRef]
  4. Rahman, A. Hecke Transport of Conifold Mixed-Hodge Extensions: Localized Survival and Higher Derived Attachments, 2026. Preprints.org 2026, version 1. [Google Scholar] [CrossRef]
  5. Rahman, A. Categorical Realizations of Hodge Atoms: Singular Spectra, Irregular Stokes, and Hecke Geometry, 2026. Preprints.org, version 1. 2026. [Google Scholar] [PubMed]
  6. Saito, M. Mixed Hodge Modules. Publ. Res. Inst. Math. Sci. 1990, 26, 221–333. [Google Scholar] [CrossRef]
  7. Saito, M. Extension of Mixed Hodge Modules. Compos. Math. 1990, 74, 209–234. [Google Scholar]
  8. Saito, M. A Young Person’s Guide to Mixed Hodge Modules. In Hodge Theory and L2-Analysis;Advanced Lectures in Mathematics; International Press: Somerville, MA, 2017; Vol. 39, pp. 517–553, [1605.00435. [Google Scholar]
  9. Schnell, C. An Overview of Morihiko Saito’s Theory of Mixed Hodge Modules. In Representation Theory, Automorphic Forms and Complex Geometry: A Tribute to Wilfried Schmid; International Press, 2020; pp. 27–80, [1405.3096. [Google Scholar]
  10. Tubach, S. On the Nori and Hodge Realisations of Voevodsky Motives. Compos. Math. 2025, 161, 2155–2201. [Google Scholar] [CrossRef]
  11. Fedorov, R. Satake Equivalence for Hodge Modules on Affine Grassmannians. arXiv 2021, arXiv:math. [Google Scholar]
  12. Achar, P.N.; Riche, S.; Vay, C. Mixed Perverse Sheaves on Flag Varieties for Coxeter Groups. Can. J. Math. 2020, 72, 1–55. [Google Scholar] [CrossRef]
  13. Lee, Y.P.; Lin, H.W.; Wang, C.L. Towards A+B Theory in Conifold Transitions for Calabi–Yau Threefolds. J. Differ. Geom. 2018, 110, 495–541. [Google Scholar] [CrossRef]
  14. He, S.; Walpuski, T. Hecke Modifications of Higgs Bundles and the Extended Bogomolny Equation. J. Geom. Phys. 2019, 146, 103487. [Google Scholar] [CrossRef]
  15. Kapustin, A.; Witten, E. Electric-Magnetic Duality and the Geometric Langlands Program. Commun. Number Theory Phys. 2007, 1, 1–236. [Google Scholar] [CrossRef]
  16. Yu, T.Y.; Zhang, S. Topological Laplace Transform and Decomposition of nc-Hodge Structures. arXiv 2024, arXiv:math. [Google Scholar]
  17. D’Agnolo, A.; Kashiwara, M. Enhanced Nearby and Vanishing Cycles in Dimension One and Fourier Transform. Publ. Res. Inst. Math. Sci. 2023, 59, 543–570. [Google Scholar] [CrossRef]
  18. Beilinson, A.A.; Bernstein, J.; Deligne, P. Faisceaux pervers; Société Mathématique de France: Astérisque, 1982; Vol. 100. [Google Scholar]
  19. MacPherson, R.; Vilonen, K. Elementary Construction of Perverse Sheaves. Invent. Math. 1986, 84, 403–435. [Google Scholar] [CrossRef]
  20. Mustaţă, M.; Popa, M. Hodge Ideals for Q-Divisors, V-Filtration, and Minimal Exponent 2020. 8. [CrossRef]
  21. Beauville, A.; Narasimhan, M.S.; Ramanan, S. Spectral Curves and the Generalised Theta Divisor. J. Für Die Reine Und Angew. Math. 1989, 398, 169–179. [Google Scholar] [CrossRef]
  22. Simpson, C.T. Higgs Bundles and Local Systems. Publ. Mathématiques De l’IHÉS 1992, 75, 5–95. [Google Scholar] [CrossRef]
  23. Donagi, R.; Pantev, T.; Simpson, C. Twistor Hecke Eigensheaves in Genus 2, 2024. arXiv arXiv:math.
  24. Beilinson, A.A. Notes on Absolute Hodge Cohomology. In Applications of Algebraic K-Theory to Algebraic Geometry and Number Theory;Contemporary Mathematics; American Mathematical Society: Providence, RI, 1986; Vol. 55, pp. 35–68. [Google Scholar] [CrossRef]
  25. Mirković, I.; Vilonen, K. Geometric Langlands Duality and Representations of Algebraic Groups over Commutative Rings. Ann. Math. 2007, 166, 95–143. [Google Scholar] [CrossRef]
  26. Witten, E. The Physical Approach to Geometric Langlands. Surv. Differ. Geom. Published in the 2015 volume; cited in subsequent 2018 literature. 2015, 20, 307–448. [Google Scholar]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.