OAI/MeasureTheory/Falconer/Angular/AngularSamples.lean
8 declarations, 0 contain sorry, 0 declared as axioms
Finite sampled direction families
Ball counts for the sampled direction labels control both multiplicities and normalized ball counts of the rounded incident line families. The estimates below retain the original labels, including repeated directions.
- def
OAI.Falconer.roundedLineFamilyline 17 - theorem
OAI.Falconer.roundedLineFamily_nonemptyline 21 - theorem
OAI.Falconer.slope_close_of_roundedLine_closeline 26Source documentation
The coordinate error from rounding is at most half the grid width.
theorem slope_close_of_roundedLine_close {δ r m : ℝ} (hδ : 0 < δ) (hδr : δ ≤ r) (x l : Space 2) (hclose : dist (roundedIncidentLine δ x m) l ≤ r) : |m - l 0| ≤ 2 * r := by …Mentions: OAI.Falconer.Space, OAI.Falconer.roundedIncidentLine
- theorem
OAI.Falconer.roundedLineFamily_ball_countline 38Source documentation
Raw ball counts for the rounded line family follow from the original direction-label counts.
theorem roundedLineFamily_ball_count {ι : Type*} [DecidableEq ι] {δ K b : ℝ} (hδ : 0 < δ) (x : Space 2) (m : ι → ℝ) (D S : Finset ι) (hSD : S ⊆ D) (hcount : ∀ a r : ℝ, δ ≤ r → ((D.filter (fun i => |m i - a| ≤ r)).card : ℝ) ≤ K * r ^ b) (l : Space 2) (r : ℝ) (hδr : δ ≤ r) : (((roundedLineFamily δ x m S).filter (fun z => dist z l ≤ r)).card : ℝ) ≤ K * 2 ^ b * r ^ b := by …Mentions: OAI.Falconer.Space, OAI.Falconer.roundedLineFamily
- theorem
OAI.Falconer.card_le_image_card_mul_fiber_boundline 67Source documentation
A finite map can lose only the maximum multiplicity of its fibers.
theorem card_le_image_card_mul_fiber_bound {ι α : Type*} [DecidableEq ι] [DecidableEq α] (S : Finset ι) (F : ι → α) {L : ℝ} (hF : ∀ a ∈ S.image F, ((S.filter (fun i => F i = a)).card : ℝ) ≤ L) : (S.card : ℝ) ≤ ((S.image F).card : ℝ) * L := by … - theorem
OAI.Falconer.roundedLineFamily_card_lowerline 76Source documentation
Even repeated original direction labels give enough distinct rounded lines: their multiplicity is controlled by one δ-ball count.
theorem roundedLineFamily_card_lower {ι : Type*} [DecidableEq ι] {δ K b : ℝ} (hδ : 0 < δ) (x : Space 2) (m : ι → ℝ) (D S : Finset ι) (hSD : S ⊆ D) (hcount : ∀ a : ℝ, ((D.filter (fun i => |m i - a| ≤ δ)).card : ℝ) ≤ K * δ ^ b) : (S.card : ℝ) ≤ ((roundedLineFamily δ x m S).card : ℝ) * (K * δ ^ b) := by …Mentions: OAI.Falconer.Space, OAI.Falconer.roundedLineFamily
- theorem
OAI.Falconer.roundedLineFamily_normalized_ball_boundline 99Source documentation
The raw sampled-family bound yields exactly the normalized finite ball bound required by the published theorem.
theorem roundedLineFamily_normalized_ball_bound {ι : Type*} [DecidableEq ι] {δ K b L : ℝ} (hδ : 0 < δ) (x : Space 2) (m : ι → ℝ) (D S : Finset ι) (hSD : S ⊆ D) (hcount : ∀ a r : ℝ, δ ≤ r → ((D.filter (fun i => |m i - a| ≤ r)).card : ℝ) ≤ K * r ^ b) (hL : K * 2 ^ b ≤ L * ((roundedLineFamily δ x m S).card : ℝ)) : NormalizedFiniteBallBound δ b L (roundedLineFamily δ x m S) := by …Mentions: OAI.Falconer.Space, OAI.Falconer.roundedLineFamily, OAI.Falconer.NormalizedFiniteBallBound
- theorem
OAI.Falconer.roundedLineFamily_normalized_of_large_sampleline 117Source documentation
If a point retains an
Ifraction of the sampled labels, the line family has normalized loss at most2 * 2^b * K^2 / I. The sample-size assumption is the one supplied byn = ceil (δ^(-b)).theorem roundedLineFamily_normalized_of_large_sample {ι : Type*} [DecidableEq ι] {δ K I b : ℝ} (hδ : 0 < δ) (hK : 0 < K) (hI : 0 < I) (x : Space 2) (m : ι → ℝ) (D S : Finset ι) (hD : D.Nonempty) (hSD : S ⊆ D) (hsize : (D.card : ℝ) * δ ^ b ≤ 2) (hretained : I * (D.card : ℝ) ≤ (S.card : ℝ)) (hcount : ∀ a r : ℝ, δ ≤ r → ((D.filter (fun i => |m i - a| ≤ r)).card : ℝ) ≤ (K * (D.card : ℝ)) * r ^ b) : NormalizedFiniteBallBound δ b (2 * 2 ^ b * K ^ 2 / I) (roundedLineFamily δ x m S) := by …Mentions: OAI.Falconer.Space, OAI.Falconer.NormalizedFiniteBallBound, OAI.Falconer.roundedLineFamily