Perimeter comparison for Harder--Narasimhan polygons #
The Harder--Narasimhan reading of the neutral perimeter comparison in
Analysis/Convex/ComplexPolygonalPath/Perimeter.lean: the vertices are
subobject charges, the edges are semistable-factor charges, and the path length
is the mass.
Main declarations #
polygonLength_le_of_vertexHull_subset, for applications that already have containment of the two closed finite vertex polygons.polygonLength_le_add_norm_charge_sub_of_mono, the boundary-cut comparison along a monomorphism, and itsmass_le_add_norm_cokernel_of_monoandmass_le_add_norm_of_shortExactmass forms.mass_eq_mass, independence of the mass of the chosen HN filtration.
Source #
Ikeda, A phase limit formula and the space of stability conditions,
Lemmas 3.7 and 3.8, at t = 0. Lemma 3.8 is strengthened at that parameter by
not requiring the cokernel to have phase one: that hypothesis controls the
exponential weight for general t and disappears when t = 0. No claim that
the full ambient HN polygon is the closed HN vertex hull is used -- it is false
in general; positive-angle support maxima of the ambient polygon come from
HNPolygon instead.
Placement #
This file is the stability adapter half of what was
StabilityCondition/Metric/Mass/Subadditivity/PolygonPerimeter.lean. MO1.13
(#1324) sent the Euclidean half to its neutral owner and gave what stayed a
name for its subject -- the HN polygon comparison -- rather than for the shape
of its proof. The citation above travels with it unchanged.
HN-path specialization of the finite perimeter comparison for applications which already have containment of the two closed finite vertex polygons. The monomorphism comparison below uses the more precise ambient support theorem instead of trying to derive this hypothesis from full ambient HN-polygon containment.
Boundary-cut HN polygon comparison. If E ⟶ E' is monic, the HN
path of E is no longer than the HN path of E' followed by the single edge
from Z(E') back to Z(E).
This is the t = 0 comparison used in Ikeda's Lemma 3.8, strengthened at that
parameter by not requiring the cokernel to have phase one. (That phase
hypothesis controls the exponential weight for general t; it disappears
when t = 0.) The final edge is allowed to lie on the positive real boundary
when the cokernel does have phase one. No false claim that the full ambient HN
polygon is the closed HN vertex hull is used. Instead, positive-angle support
maxima of the ambient polygon are supplied by HNPolygon and the finite
support-fan perimeter theorem closes the argument.
Mass form of the boundary-cut comparison. The closing-edge charge is the negative of the cokernel charge, by additivity of the stability function.
Short-exact-sequence form of the boundary-cut comparison.
The mass of an abelian HN filtration is independent of the chosen filtration. This is the identity-monomorphism specialization of the boundary-cut comparison.