The two classes at a weak slope cutoff #
Cutoff.lean cuts two classes out of an abelian category by a phase
β : ℝ, which is available for a strict stability function. This file cuts
them by a slope μ₀ : WithTop ℝ for a weak one:
T μ₀ = {E | every weak HN slope of E exceeds μ₀}
F μ₀ = {E | every weak HN slope of E is at most μ₀}
Why the cutoff is WithTop ℝ, and what that buys #
Because slopes are. A rank-zero object has charge on the real boundary, so
slope ⊤, so μMinus = ⊤ > μ₀ for every finite μ₀, so it lies in
T μ₀ — mem_hnTors_of_slope_eq_top below. That is the classical convention:
torsion sheaves are in the torsion class at every finite slope cutoff.
This is the correct replacement for SlopeCutoff.lean's
mem_hnTors_of_rank_zero, and the replacement is not cosmetic. The strict
lemma reaches its conclusion through phase_eq_one_of_rank_zero, which needs
degree_pos_of_rank_zero — the curve hypothesis, false on a surface, where
a skyscraper has c₁ = 0 and so degree 0. With charge exactly 0 there is
no phase at all (arg 0 = 0 is not a phase), and no phase comparison can place
the object. The slope ⊤ places it, with no positivity hypothesis whatever.
The cutoff μ₀ = ⊤ is permitted and degenerate in the expected direction: no
slope exceeds ⊤, so T ⊤ contains only the zero object while F ⊤ is
everything. Nothing below excludes it.
What is here and what is next #
hom_eq_zero_of_mem_hnTors_of_mem_hnFree is the Hom-vanishing half of a torsion
pair. The splitting half is WeakSplitting.lean, and with both present the two
classes are a torsion pair — assembled as
WeakStabilityFunctionOn.hnTorsionPair in
Weak/Tilting/TorsionPair/WeakStabilityFunction.lean. This file still states
only the Hom-vanishing, which is all it needs.
The torsion class at a weak slope cutoff: the zero object, and the
objects whose weak HN slopes all exceed μ₀.
Equations
Instances For
The torsion-free class at a weak slope cutoff: the zero object, and the
objects whose weak HN slopes are all at most μ₀.
Equations
Instances For
Membership is independent of the filtration chosen, because μ⁻ is.
The HN property is needed for the reverse direction only: without a filtration to exhibit, the universally quantified condition is vacuous while membership asks for a witness.
Membership is independent of the filtration chosen, because μ⁺ is.
A weak-semistable object of slope ⊤ lies in the torsion class at every
finite cutoff. This is the skyscraper, and it is the whole reason the cutoff
is valued in WithTop ℝ rather than in ℝ.
A weak-semistable object of charge 0 lies in the torsion class at every
finite cutoff. The rank-zero, degree-zero case in charge terms.
The two classes meet only in zero. An object in both would have
μ₀ < μ⁻ ≤ μ⁺ ≤ μ₀.
Hom-vanishing, for a weak-semistable target. A map from an object whose
weak HN slopes all exceed μ₀ to a weak-semistable object of slope at most
μ₀ is zero.
Hom-vanishing. A map from an object whose weak HN slopes all exceed
μ₀ to an object whose weak HN slopes are all at most μ₀ is zero: its image
is a quotient of the source and a subobject of the target, so
μ₀ < μ⁻(image) and μ⁺(image) ≤ μ₀, which μMinus_le_μPlus contradicts
unless the image is zero.
The HN property is needed for the image, which is neither of the two given objects.