Grassmannian of unoriented planes
DefinitionOpenGA_UnorientedGrassmanniangeometric-measure-theoryopenga-varifoldpoincare-foundationsvarifold
For a finite-dimensional real inner-product space , consists of its exactly -dimensional linear subspaces, without orientation. Its distance is the operator norm of the difference of the orthogonal projections. This compact metric space has its Borel measurable structure. If , it is empty. The compactness and projection lemmas here support the measure constructions. Simon uses the Euclidean matrix norm; this formalization uses the operator norm on the same finite-dimensional operator space.
Definition code
import Mathlib.Analysis.InnerProductSpace.Adjoint
import Mathlib.Analysis.InnerProductSpace.Projection.FiniteDimensional
import Mathlib.LinearAlgebra.Trace
import Mathlib.Topology.MetricSpace.Isometry
import Mathlib.Topology.Algebra.Module.ContinuousLinearMap.Idempotent
import Mathlib.MeasureTheory.Constructions.BorelSpace.Basic
/-!
# The Grassmannian of unoriented planes
`OpenGA.Grassmannian E k` consists of exactly the `k`-dimensional real subspaces of `E`.
The metric is the operator norm of the difference of orthogonal projections. This is the
projection model of the Grassmannian; no orientation or choice of basis is included.
Reference: Leon Simon, *Introduction to Geometric Measure Theory*, 2018 lecture notes,
Chapter 8, Section 1, pp. 235-236. Simon uses the Euclidean matrix norm of projections;
we use the operator norm, another norm on the same finite-dimensional operator space.
The compactness proof below uses the closed, bounded set of self-adjoint idempotents
of trace `k`. This is distinct from Mathlib's algebraic, quotient-rank Grassmannian.
-/
noncomputable section
open Set Topology
namespace OpenGA
/-- Unoriented `k`-planes in a finite-dimensional real inner product space. -/
def Grassmannian (E : Type*) [NormedAddCommGroup E] [InnerProductSpace ℝ E]
[FiniteDimensional ℝ E] (k : ℕ) :=
{S : Submodule ℝ E // Module.finrank ℝ S = k}
namespace Grassmannian
variable {E : Type*} [NormedAddCommGroup E] [InnerProductSpace ℝ E]
[FiniteDimensional ℝ E] {k : ℕ}
/-- The underlying plane. Its dimension is part of the definition. -/
def plane (P : Grassmannian E k) : Submodule ℝ E := P.val
@[simp] theorem finrank_plane (P : Grassmannian E k) :
Module.finrank ℝ P.plane = k := P.property
@[ext] theorem ext {P Q : Grassmannian E k} (h : P.plane = Q.plane) : P = Q :=
Subtype.ext h
/-- Orthogonal projection onto the plane, with values in the ambient space. -/
def projection (P : Grassmannian E k) : E →L[ℝ] E := P.plane.starProjection
theorem projection_injective : Function.Injective (projection : Grassmannian E k → E →L[ℝ] E) := by
intro P Q h
apply ext
have hr := congrArg (fun A : E →L[ℝ] E => A.range) h
simpa [projection, Submodule.range_starProjection] using hr
instance : MetricSpace (Grassmannian E k) :=
MetricSpace.induced projection projection_injective inferInstance
theorem isometry_projection : Isometry (projection : Grassmannian E k → E →L[ℝ] E) :=
fun _ _ => rfl
theorem continuous_projection : Continuous (projection : Grassmannian E k → E →L[ℝ] E) :=
isometry_projection.continuous
@[simp] theorem dist_eq (P Q : Grassmannian E k) :
dist P Q = ‖P.projection - Q.projection‖ :=
(isometry_projection.dist_eq P Q).symm.trans (dist_eq_norm _ _)
instance : SecondCountableTopology (Grassmannian E k) :=
isometry_projection.isEmbedding.secondCountableTopology
instance : MeasurableSpace (Grassmannian E k) := borel _
instance : BorelSpace (Grassmannian E k) := ⟨rfl⟩
theorem projection_mem (P : Grassmannian E k) (v : E) : P.projection v ∈ P.plane :=
P.plane.starProjection_apply_mem v
@[simp] theorem projection_eq_self_iff (P : Grassmannian E k) (v : E) :
P.projection v = v ↔ v ∈ P.plane := P.plane.starProjection_eq_self_iff
theorem norm_projection_le (P : Grassmannian E k) : ‖P.projection‖ ≤ 1 :=
P.plane.starProjection_norm_le
theorem isStarProjection_projection (P : Grassmannian E k) : IsStarProjection P.projection :=
isStarProjection_starProjection
theorem trace_projection (P : Grassmannian E k) :
LinearMap.trace ℝ E P.projection.toLinearMap = k := by
have hid := ContinuousLinearMap.IsIdempotentElem.toLinearMap
P.plane.isIdempotentElem_starProjection
have h := (LinearMap.IsIdempotentElem.isProj_range _ hid).trace
change LinearMap.trace ℝ E P.projection.toLinearMap =
(Module.finrank ℝ P.plane.starProjection.range : ℝ) at h
rw [Submodule.range_starProjection, P.finrank_plane] at h
exact h
/-- The image of the projection representation is described by closed algebraic conditions. -/
theorem range_projection :
Set.range (projection : Grassmannian E k → E →L[ℝ] E) =
{A | IsStarProjection A ∧ LinearMap.trace ℝ E A.toLinearMap = k} := by
ext A
constructor
· rintro ⟨P, rfl⟩
exact ⟨P.isStarProjection_projection, P.trace_projection⟩
· rintro ⟨hA, htr⟩
obtain ⟨_, heq⟩ := isStarProjection_iff_eq_starProjection_range.mp hA
have hdim : Module.finrank ℝ A.range = k := by
have hid := ContinuousLinearMap.IsIdempotentElem.toLinearMap hA.isIdempotentElem
have ht := (LinearMap.IsIdempotentElem.isProj_range _ hid).trace
exact_mod_cast ht.symm.trans htr
exact ⟨⟨A.range, hdim⟩, heq.symm⟩
theorem isClosed_range_projection :
IsClosed (Set.range (projection : Grassmannian E k → E →L[ℝ] E)) := by
rw [range_projection]
have htrace : Continuous (fun A : E →L[ℝ] E => LinearMap.trace ℝ E A.toLinearMap) :=
((LinearMap.trace ℝ E).comp (ContinuousLinearMap.coeLM ℝ)).continuous_of_finiteDimensional
have hid : IsClosed {A : E →L[ℝ] E | A * A = A} :=
isClosed_eq (continuous_id.mul continuous_id) continuous_id
have hstar : IsClosed {A : E →L[ℝ] E | star A = A} :=
isClosed_eq continuous_star continuous_id
simpa only [isStarProjection_iff, IsIdempotentElem, IsSelfAdjoint, Set.ofPred_and] using
(hid.inter hstar).inter (isClosed_eq htrace (continuous_const (y := (k : ℝ))))
/-- The Grassmannian is compact, including the empty case `k > dim E`. -/
instance : CompactSpace (Grassmannian E k) := by
have hbounded : Bornology.IsBounded
(Set.range (projection : Grassmannian E k → E →L[ℝ] E)) := by
apply (Metric.isBounded_closedBall (x := (0 : E →L[ℝ] E)) (r := 1)).subset
rintro A ⟨P, rfl⟩
simpa using P.norm_projection_le
have hc := Metric.isCompact_of_isClosed_isBounded isClosed_range_projection hbounded
exact ⟨isometry_projection.isEmbedding.isCompact_iff.mpr (by simpa using hc)⟩
end Grassmannian
end OpenGA
Source