Where the rank-zero objects go #
Slope.lean puts a nonzero object of rank zero on the negative real axis, by
degree_pos_of_rank_zero. This file draws the consequence for the cut of
Cutoff.lean: such an object has phase one, the largest a phase can be, and
so lies in the torsion class T β for every cutoff β < 1.
That is not a convention chosen here. It is forced by the charge
-degree + i · rank, and it is what makes the geometric statement come out
right: on a surface the rank-zero coherent sheaves are the torsion sheaves, and
Bridgeland's T β contains all of them at every β.
Why the Harder–Narasimhan filtration cannot get in the way #
Membership in T β asks for β < φ⁻, not for semistability, so the argument
has to reach the last HN factor of a rank-zero object. It is again rank zero:
rank is additive and nonnegative, so a subobject and its quotient both inherit
rank zero from the whole, and the last HN factor is such a quotient. Hence its
phase is one, hence φ⁻ = 1.
Rank zero passes to both ends of a short exact sequence, because rank is additive and never negative.
A nonzero object of rank zero has phase one. Its charge is
-degree, a negative real, and arg of a negative real is π.
Rank-zero objects are torsion at every cutoff below one. The last
Harder–Narasimhan factor of a rank-zero object again has rank zero, so φ⁻ is
one.