Regularization models and strictly convex star-shaped points
DefinitionBirkhoffGlobalSection_RegularizationModelRegularization models of the selected Levi-Civita component, and strictly convex star-shaped points of a level set.
Let be the selected Levi-Civita component, with Levi-Civita Hamiltonian and Hamiltonian vector field . A regularization model of consists of the following data.
- Open sets and in , and mutually inverse smooth maps and .
- A constant with for all , where .
- Antipodal equivariance on the component: for .
- A smooth model Hamiltonian on with and for .
- A continuous function on , positive on , with
So carries the Levi-Civita flow on to the model flow on up to a positive change of time. These are the fields of ConvexRegularizationModel without the convex body. The notion covers other regularizations of the same component, such as the elliptic-hyperbolic coordinates used by Liu–Salomão, even where the model surface is convex only on part of the component.
A point is a strictly convex star-shaped point of if is near and
Geometrically, the level set of through is transverse to the ray through , with increasing outward, and its second fundamental form is positive definite at . This is the pointwise version of the strict convexity and star-shapedness used for dynamically convex energy surfaces.
Formalization Note The model is RegularizationModel μ c, with fields toModel, fromModel, symplecticScale, modelHamiltonian and timeScale. The pointwise predicate is IsStrictlyConvexStarShapedAt G y, defined as HasPositiveTangentialHessianOn G {y} together with 0 < fderiv ℝ G y y. The symplectic form is phaseSymplecticForm from Def_BirkhoffGlobalSection_ConvexRegularizationModel.
import Definitions.Def_BirkhoffGlobalSection_ConvexRegularizationModel
namespace BirkhoffGlobalSection
open scoped ContDiff
/-- A smooth change of regularized coordinates for the selected Levi-Civita
component, with no convexity requirement.
The coordinate maps are smooth inverses on open neighborhoods, commute with the
antipodal map on the component, and preserve the symplectic form up to a
positive constant. The component lands in the regular zero level of a model
Hamiltonian, and the Levi-Civita field is carried to a positive multiple of the
model Hamiltonian field, including at collision points. The fields are those of
`ConvexRegularizationModel` without the convex body, so an alternative
regularization such as elliptic-hyperbolic coordinates can be recorded even
where it is only partially convex. -/
structure RegularizationModel (μ c : ℝ) where
source : Set Phase
target : Set Phase
source_open : IsOpen source
target_open : IsOpen target
contains_component : leftEnergyComponent μ c ⊆ source
toModel : Phase → Phase
fromModel : Phase → Phase
toModel_mapsTo : Set.MapsTo toModel source target
fromModel_mapsTo : Set.MapsTo fromModel target source
left_inverse : ∀ s ∈ source, fromModel (toModel s) = s
right_inverse : ∀ y ∈ target, toModel (fromModel y) = y
toModel_smooth : ContDiffOn ℝ ∞ toModel source
fromModel_smooth : ContDiffOn ℝ ∞ fromModel target
symplecticScale : ℝ
symplecticScale_pos : 0 < symplecticScale
symplectic : ∀ s ∈ source, ∀ v w : Phase,
phaseSymplecticForm (fderiv ℝ toModel s v) (fderiv ℝ toModel s w) =
symplecticScale * phaseSymplecticForm v w
antipodal : ∀ s ∈ leftEnergyComponent μ c, toModel (-s) = -toModel s
modelHamiltonian : Phase → ℝ
model_smooth : ContDiffOn ℝ ∞ modelHamiltonian target
model_zero : ∀ s ∈ leftEnergyComponent μ c, modelHamiltonian (toModel s) = 0
model_regular : ∀ s ∈ leftEnergyComponent μ c,
fderiv ℝ modelHamiltonian (toModel s) ≠ 0
timeScale : Phase → ℝ
timeScale_continuous : ContinuousOn timeScale source
timeScale_pos : ∀ s ∈ leftEnergyComponent μ c, 0 < timeScale s
vectorField : ∀ s ∈ leftEnergyComponent μ c,
fderiv ℝ toModel s (hamiltonianVectorField (leviCivitaHamiltonian μ c) s) =
timeScale s • hamiltonianVectorField modelHamiltonian (toModel s)
/-- The level set of `G` through `y` is strictly convex and star-shaped at `y`:
`G` is `C²` near `y`, its Hessian is positive on nonzero vectors annihilated by
`dG_y`, and `dG_y(y) > 0`, so the radial direction leaves the level set toward
increasing `G`. -/
def IsStrictlyConvexStarShapedAt (G : Phase → ℝ) (y : Phase) : Prop :=
HasPositiveTangentialHessianOn G {y} ∧ 0 < fderiv ℝ G y y
end BirkhoffGlobalSection