The support property and its quadratic form #
Kontsevich–Soibelman's support property for a central charge asks for a
uniform comparison ‖v‖ ≤ C ‖Z v‖ over a distinguished set of classes. The
formulation used throughout the modern literature replaces that estimate by a
quadratic form Q which is nonnegative on the distinguished set and
negative definite on ker Z. This file proves the two are equivalent.
What this is a theorem about #
S here is an arbitrary subset of a finite-dimensional real normed space.
In the intended application it is the set of classes of σ-semistable objects,
but that identification needs a stability condition and is not made here: no
declaration below mentions a triangulated category, a slicing, or anything
else in StabilityCondition/. Every statement is linear algebra plus one
compactness argument, and
would hold verbatim for any other choice of S.
The intended W is ℂ; the proofs never use anything about the target beyond
its norm, so it is left general.
Why the hypotheses on Q are what they are #
The proofs use exactly two properties of a quadratic form: continuity and
degree-two homogeneity. Both directions are therefore stated for
IsHomogTwo, which is strictly weaker than being a quadratic form — see its
docstring. This makes both theorems stronger and costs nothing, since every
quadratic form on a finite-dimensional real space satisfies it.
Main results #
hasSupportProperty_of_isCompatible— the compactness direction.exists_isCompatible_of_hasSupportProperty— the explicit direction, viaQ v = C² ‖Z v‖² - ‖v‖².hasSupportProperty_iff— the equivalence.
The support property: a uniform linear comparison of the norm of a class
with the norm of its charge, over the distinguished set S.
Equations
Instances For
The two properties of a quadratic form that the equivalence below actually uses: continuity and homogeneity of degree two.
This is deliberately weaker than being a quadratic form. Continuity plus
Q (a • v) = a² Q v does not imply bilinearity: Q (x, y) = |x * y| on ℝ²
satisfies both and is not a quadratic form. Stating the hypothesis this way
makes both directions stronger theorems and costs nothing, because every
quadratic form on a finite-dimensional real space satisfies it.
- continuous : Continuous Q
Qis continuous Qis homogeneous of degree two
Instances For
Compatibility of a form with the pair (Z, S): nonnegative on S, and
negative definite on ker Z.
Qis nonnegative on the distinguished setQis negative definite on the kernel of the charge
Instances For
The compact slice the argument runs on: the unit sphere cut by Q ≥ 0.
Equations
- CategoryTheory.Triangulated.WeakStabilityCondition.Support.slice Q = Metric.sphere 0 1 ∩ {v : V | 0 ≤ Q v}
Instances For
Normalising a nonzero vector on which Q is nonnegative lands in the
slice. This is the only place degree-two homogeneity is used.
The support property follows from a compatible form.
The slice is compact and Z does not vanish on it, so ‖Z ·‖ attains a
positive minimum there; the comparison constant is its reciprocal.
A compatible form follows from the support property.
The witness is explicit: Q v = C² ‖Z v‖² - ‖v‖². On S it is nonnegative by
squaring the support estimate, and on ker Z it is -‖v‖².
The support property is exactly the existence of a compatible form.
This is the Kontsevich–Soibelman reformulation, at the level of the numerical data it is a statement about.
Elementary consequences #
Both are immediate from the definition and are stated because they are the
sanity checks a reader will want: the property is monotone in S, and it
forces the charge to be injective on the span of S in the strong pointwise
sense.
A class in S with vanishing charge is zero.
Stability under perturbation of the charge #
The support property is an open condition on Z. That is what makes it the
right hypothesis for a deformation argument: a charge that has it keeps it under
any sufficiently small change, so the property propagates along a path rather
than having to be re-established at every point.
None of this uses compactness, the quadratic form, or finite-dimensionality —
only the triangle inequality and one division — so all three are stated with
FiniteDimensional omitted. The topological statement at the end is the only
one that needs a norm on the space of charges, and it is a corollary of the
other two rather than a separate argument.
The support property survives a small perturbation of the charge.
Both the constant and the closeness bound are explicit, because the admissible
size of the perturbation depends on C: an estimate that is only barely true
tolerates only a barely nonzero change. The perturbed constant is
C / (1 - Cε), which degrades to +∞ exactly as Cε → 1.
Openness, stated without a topology on the space of charges.
Every charge with the support property admits a positive tolerance inside which
every other charge still has it. isOpen_hasSupportProperty is this made
topological; this form is separated out because it needs no norm on the space of
maps and so applies to plain linear maps.
The charges with the support property form an open set.
S is fixed and the charges are topologized by the operator norm. This is the
IsOpen form of HasSupportProperty.exists_tolerance; the mathematical content
is entirely in that lemma, and the only step here is le_opNorm.