The dual involution on a numerical intersection ring #
(-)^∨ acts on the codimension-i part of A^•(X)_ℚ by (-1)^i. It is what turns
Hirzebruch–Riemann–Roch into its bilinear form χ(E,F) = ∫_X ch(E)^∨ · ch(F) · td(X), and
so it is the last piece of Layer A that Bridgeland stability needs: the central charge
factors through N(X) modulo the radical of that pairing.
Why a separate data bundle #
NumericalRingDualData R is an explicit extension of one selected presentation
R : NumericalRingData n A, not a field of every numerical ring and not an independent
bundle with another path back to R.
A ring that has no chosen dual needs no placeholder, while two choices can coexist without
instance search selecting between them.
Main definitions #
NumericalRingDualData R— the involution, as data plus the axiom that it acts by(-1)^ionpiece i.chDual E—ch(E)^∨, written directly asΣᵢ (-1)ⁱ chᵢ(E)rather than asdual (ch E).
Main results #
dual_ch— the two descriptions ofch(E)^∨agree.dual_involutive_of_mem_piece—(-)^∨is an involution on each graded piece.
Deliberately not stated #
dual (dual x) = x for arbitrary x, and degree (dual x) = (-1)^n · degree x. Both are
true, and both need the graded decomposition of x — NumericalRingData.isInternal — rather
than just membership in a single piece. Nothing downstream needs them: every consumer here
reaches dual through chComp and toddComp, which are graded by construction, so the
component-wise lemmas suffice. Adding the general versions means importing
DirectSum.IsInternal machinery for no current gain.
The dual involution on a numerical intersection ring: a ℚ-algebra endomorphism
acting on codimension i by (-1)^i.
Geometrically this is E ↦ E^∨ read through the Chern character: chᵢ(E^∨) = (-1)ⁱchᵢ(E).
It is carried as data because NumericalRingData does not expose the graded projections, so
there is nothing to define it from at this layer.
The involution, as a
ℚ-algebra map.It acts on codimension
iby(-1)^i.
Instances For
(-)^∨ preserves each graded piece: it only rescales it.
(-)^∨ is an involution on each graded piece.
Scalars are fixed by (-)^∨, since they sit in codimension zero.
The Chern character is additive, componentwise additivity summed up.
Stated before the dual is in scope because it has nothing to do with it.
The dual Chern character ch(E)^∨ = Σᵢ (-1)ⁱ chᵢ(E).
Defined from the grading alone, with no NumericalRingDualData argument in sight: the
alternating sum is available on any NumericalVarietyData, and dual_ch below is what says it
deserves the name. Everything the Euler pairing needs lives at this level of generality.
Instances For
chᵢ(E^∨) = (-1)ⁱ chᵢ(E).
tdᵢ(X)^∨ = (-1)ⁱ tdᵢ(X).
ch(E)^∨ really is ch(E) pushed through the involution.