math / data /lean /OAI /Probability /FreeIsing /Spectral /OrthogonalRowIdentification.lean
introvoyz045's picture
Migrated from GitHub
4a56cd6 verified
Raw History Blame Contribute Delete
3.45 kB
import OAI.Probability.FreeIsing.Posterior.FilteringRepresentationClosure
namespace OAI
noncomputable section
open MeasureTheory Filter
open scoped Topology
namespace Problem367.FilteringRepresentationClosure
variable {E : Type*} [NormedAddCommGroup E] [InnerProductSpace ℝ E]
theorem inner_sum_coordinate {ΞΉ : Type*} {f : ΞΉ β†’ E} {y : E}
(hs : HasSum f y) (ho : Pairwise fun i j => inner ℝ (f i) (f j) = 0) (i : ΞΉ) :
inner ℝ y (f i) = β€–f iβ€– ^ 2 := by
have h := hs.mapL (innerSL ℝ (f i))
change HasSum (fun j => inner ℝ (f i) (f j)) (inner ℝ (f i) y) at h
have hsingle : HasSum (fun j => inner ℝ (f i) (f j)) (inner ℝ (f i) (f i)) :=
hasSum_single i (fun j hji => ho hji.symm)
rw [real_inner_comm, ← real_inner_self_eq_norm_sq]
exact h.unique hsingle
theorem eq_of_orthogonal_row_limits
{ΞΉ N : Type*} {l : Filter N} [l.NeBot]
{f : N β†’ ΞΉ β†’ E} {y : N β†’ E} {g : ΞΉ β†’ E} {z m : E}
(hs : βˆ€ n, HasSum (f n) (y n))
(ho : βˆ€ n, Pairwise fun i j => inner ℝ (f n i) (f n j) = 0)
(hf : βˆ€ i, Tendsto (fun n => f n i) l (𝓝 (g i)))
(hy : Tendsto y l (𝓝 z)) (hg : HasSum g m)
(he : β€–zβ€– ^ 2 = β€–mβ€– ^ 2) : z = m := by
have hgo : Pairwise fun i j => inner ℝ (g i) (g j) = 0 := by
intro i j hij
have h := (hf i).inner (π•œ := ℝ) (hf j)
have hz : Tendsto (fun n => inner ℝ (f n i) (f n j)) l (𝓝 (0 : ℝ)) := by
simpa only [ho _ hij] using (tendsto_const_nhds : Tendsto (fun _ : N => (0 : ℝ)) l (𝓝 0))
exact tendsto_nhds_unique h hz
have hcross (i : ΞΉ) : inner ℝ z (g i) = β€–g iβ€– ^ 2 := by
have hleft := hy.inner (π•œ := ℝ) (hf i)
have hright := (hf i).norm.pow 2
exact tendsto_nhds_unique_of_eventuallyEq hleft hright
(Eventually.of_forall fun n => inner_sum_coordinate (hs n) (ho n) i)
have hzi := hg.mapL (innerSL ℝ z)
have hmi := hg.mapL (innerSL ℝ m)
change HasSum (fun i => inner ℝ z (g i)) (inner ℝ z m) at hzi
change HasSum (fun i => inner ℝ m (g i)) (inner ℝ m m) at hmi
simp only [hcross] at hzi
simp only [inner_sum_coordinate hg hgo] at hmi
exact eq_of_cross_energy (by simpa only [real_inner_self_eq_norm_sq] using hzi.unique hmi) he
theorem ae_eq_of_orthogonal_row_meanSquare_limits
{Ξ© ΞΉ N : Type*} [MeasurableSpace Ξ©] {P : Measure Ξ©}
{l : Filter N} [l.NeBot]
{X : N β†’ ΞΉ β†’ Ξ© β†’ ℝ} {Y : N β†’ Ξ© β†’ ℝ} {I : ΞΉ β†’ Ξ© β†’ ℝ} {Z M : Ξ© β†’ ℝ}
(hX : βˆ€ n i, MemLp (X n i) 2 P) (hY : βˆ€ n, MemLp (Y n) 2 P)
(hI : βˆ€ i, MemLp (I i) 2 P) (hZ : MemLp Z 2 P) (hM : MemLp M 2 P)
(hs : βˆ€ n, HasSum (fun i => (hX n i).toLp (X n i)) ((hY n).toLp (Y n)))
(ho : βˆ€ n, Pairwise fun i j => ∫ Ο‰, X n i Ο‰ * X n j Ο‰ βˆ‚P = 0)
(hc : βˆ€ i, Tendsto (fun n => ∫ Ο‰, (X n i Ο‰ - I i Ο‰) ^ 2 βˆ‚P) l (𝓝 0))
(hy : Tendsto (fun n => ∫ Ο‰, (Y n Ο‰ - Z Ο‰) ^ 2 βˆ‚P) l (𝓝 0))
(hi : HasSum (fun i => (hI i).toLp (I i)) (hM.toLp M))
(he : (∫ Ο‰, Z Ο‰ ^ 2 βˆ‚P) = ∫ Ο‰, M Ο‰ ^ 2 βˆ‚P) : Z =ᡐ[P] M := by
apply (hZ.toLp_eq_toLp_iff hM).mp
apply eq_of_orthogonal_row_limits hs
(fun n => by simpa only [PredictableL2.inner_toLp] using ho n)
(fun i => tendsto_toLp_of_integral_sq_error (fun n => hX n i) (hI i) (hc i))
(tendsto_toLp_of_integral_sq_error hY hZ hy) hi
simpa only [PredictableL2.norm_toLp_sq] using he
end Problem367.FilteringRepresentationClosure
end
end OAI