math / data /lean /OAI /Probability /FreeIsing /Spectral /SpectralClosure.lean
introvoyz045's picture
Migrated from GitHub
4a56cd6 verified
Raw History Blame Contribute Delete
6.29 kB
import OAI.Probability.FreeIsing.Spectral.SpectralTransfer
namespace OAI
noncomputable section
namespace Problem367
namespace SpectralClosure
variable {๐•œ : Type*} [RCLike ๐•œ]
variable {E F : Type*}
[NormedAddCommGroup E] [NormedSpace ๐•œ E]
[NormedAddCommGroup F] [InnerProductSpace ๐•œ F] [CompleteSpace F]
lemma norm_apply_le_on_closure_range
(T : E โ†’L[๐•œ] E) (S : F โ†’L[๐•œ] F) (D : E โ†’L[๐•œ] F)
(hS : IsSelfAdjoint S) (hDS : โˆ€ x, S (D x) = D (T x))
{ฯ : โ„} (hฯ : 0 โ‰ค ฯ) (hT : โ€–Tโ€– โ‰ค ฯ)
{y : F} (hy : y โˆˆ closure (Set.range D)) : โ€–S yโ€– โ‰ค ฯ * โ€–yโ€– :=
selfAdjoint_norm_apply_le_on_closure_range_of_intertwiner S hS T D hDS hฯ hT hy
def rangeClosure (D : E โ†’L[๐•œ] F) : Submodule ๐•œ F :=
(LinearMap.range (D : E โ†’โ‚—[๐•œ] F)).topologicalClosure
instance rangeClosure_completeSpace (D : E โ†’L[๐•œ] F) : CompleteSpace (rangeClosure D) :=
(Submodule.isClosed_topologicalClosure (LinearMap.range (D : E โ†’โ‚—[๐•œ] F))).isComplete.completeSpace_coe
omit [CompleteSpace F] in
@[simp] lemma mem_rangeClosure (D : E โ†’L[๐•œ] F) (y : F) :
y โˆˆ rangeClosure D โ†” y โˆˆ closure (Set.range D) := Iff.rfl
omit [CompleteSpace F] in
lemma mapsTo_rangeClosure (T : E โ†’L[๐•œ] E) (S : F โ†’L[๐•œ] F) (D : E โ†’L[๐•œ] F)
(hDS : โˆ€ x, S (D x) = D (T x)) :
โˆ€ y โˆˆ rangeClosure D, S y โˆˆ rangeClosure D := by
have hmap : Set.MapsTo S (Set.range D) (Set.range D) := by
rintro y โŸจx, rflโŸฉ
exact โŸจT x, (hDS x).symmโŸฉ
exact hmap.closure S.continuous
def restrict (T : E โ†’L[๐•œ] E) (S : F โ†’L[๐•œ] F) (D : E โ†’L[๐•œ] F)
(hDS : โˆ€ x, S (D x) = D (T x)) : rangeClosure D โ†’L[๐•œ] rangeClosure D :=
S.restrict (mapsTo_rangeClosure T S D hDS)
lemma restrict_isSymmetric (T : E โ†’L[๐•œ] E) (S : F โ†’L[๐•œ] F) (D : E โ†’L[๐•œ] F)
(hS : IsSelfAdjoint S) (hDS : โˆ€ x, S (D x) = D (T x)) :
((restrict T S D hDS) : rangeClosure D โ†’โ‚—[๐•œ] rangeClosure D).IsSymmetric := by
intro x y
exact hS.isSymmetric x y
lemma restrict_isSelfAdjoint (T : E โ†’L[๐•œ] E) (S : F โ†’L[๐•œ] F) (D : E โ†’L[๐•œ] F)
(hS : IsSelfAdjoint S) (hDS : โˆ€ x, S (D x) = D (T x)) :
IsSelfAdjoint (restrict T S D hDS) :=
(restrict_isSymmetric T S D hS hDS).isSelfAdjoint
lemma mapsTo_orthogonal_rangeClosure
(T : E โ†’L[๐•œ] E) (S : F โ†’L[๐•œ] F) (D : E โ†’L[๐•œ] F)
(hS : IsSelfAdjoint S) (hDS : โˆ€ x, S (D x) = D (T x)) :
โˆ€ y โˆˆ (rangeClosure D)แ—ฎ, S y โˆˆ (rangeClosure D)แ—ฎ := by
intro y hy
rw [Submodule.mem_orthogonal]
intro x hx
calc
inner ๐•œ x (S y) = inner ๐•œ (S x) y := by
simpa only [ContinuousLinearMap.coe_coe] using (hS.isSymmetric x y).symm
_ = 0 := Submodule.inner_right_of_mem_orthogonal (mapsTo_rangeClosure T S D hDS x hx) hy
lemma norm_restrict_le (T : E โ†’L[๐•œ] E) (S : F โ†’L[๐•œ] F) (D : E โ†’L[๐•œ] F)
(hS : IsSelfAdjoint S) (hDS : โˆ€ x, S (D x) = D (T x))
{ฯ : โ„} (hฯ : 0 โ‰ค ฯ) (hT : โ€–Tโ€– โ‰ค ฯ) : โ€–restrict T S D hDSโ€– โ‰ค ฯ := by
apply ContinuousLinearMap.opNorm_le_bound _ hฯ
intro y
exact norm_apply_le_on_closure_range T S D hS hDS hฯ hT y.property
omit [CompleteSpace F] in
lemma range_domRestrict_eq (D : E โ†’L[๐•œ] F) (K : Submodule ๐•œ E)
(hcover : โˆ€ x : E, โˆƒ z โˆˆ K, D z = D x) :
Set.range (D.domRestrict K) = Set.range D := by
ext y
constructor
ยท rintro โŸจx, rflโŸฉ
exact โŸจx, rflโŸฉ
ยท rintro โŸจx, rflโŸฉ
obtain โŸจz, hz, heqโŸฉ := hcover x
exact โŸจโŸจz, hzโŸฉ, heqโŸฉ
omit [CompleteSpace F] in
lemma rangeClosure_domRestrict_eq (D : E โ†’L[๐•œ] F) (K : Submodule ๐•œ E)
(hcover : โˆ€ x : E, โˆƒ z โˆˆ K, D z = D x) :
rangeClosure (D.domRestrict K) = rangeClosure D := by
ext y
simp only [mem_rangeClosure, range_domRestrict_eq D K hcover]
omit [CompleteSpace F] in
lemma source_cover_of_kills_direction (D : E โ†’L[๐•œ] F) (โ„“ : E โ†’L[๐•œ] ๐•œ)
(e : E) (he : โ„“ e = 1) (hDe : D e = 0) :
โˆ€ x : E, โˆƒ z โˆˆ LinearMap.ker (โ„“ : E โ†’โ‚—[๐•œ] ๐•œ), D z = D x := by
intro x
refine โŸจx - โ„“ x โ€ข e, ?_, ?_โŸฉ
ยท change โ„“ (x - โ„“ x โ€ข e) = 0
simp [he]
ยท simp [hDe]
omit [CompleteSpace F] in
lemma rangeClosure_ker_eq_of_kills_direction (D : E โ†’L[๐•œ] F) (โ„“ : E โ†’L[๐•œ] ๐•œ)
(e : E) (he : โ„“ e = 1) (hDe : D e = 0) :
rangeClosure (D.domRestrict (LinearMap.ker (โ„“ : E โ†’โ‚—[๐•œ] ๐•œ))) = rangeClosure D :=
rangeClosure_domRestrict_eq D _ (source_cover_of_kills_direction D โ„“ e he hDe)
lemma norm_apply_le_of_source_subspace
(T : E โ†’L[๐•œ] E) (S : F โ†’L[๐•œ] F) (D : E โ†’L[๐•œ] F)
(K : Submodule ๐•œ E) (hK : โˆ€ x โˆˆ K, T x โˆˆ K)
(hS : IsSelfAdjoint S) (hDS : โˆ€ x, S (D x) = D (T x))
(hcover : โˆ€ x : E, โˆƒ z โˆˆ K, D z = D x)
{ฯ : โ„} (hฯ : 0 โ‰ค ฯ) (hT : โ€–T.restrict hKโ€– โ‰ค ฯ)
{y : F} (hy : y โˆˆ closure (Set.range D)) : โ€–S yโ€– โ‰ค ฯ * โ€–yโ€– := by
apply norm_apply_le_on_closure_range (T.restrict hK) S (D.domRestrict K) hS
(fun x => hDS x) hฯ hT
rwa [range_domRestrict_eq D K hcover]
lemma norm_restrict_le_of_source_subspace
(T : E โ†’L[๐•œ] E) (S : F โ†’L[๐•œ] F) (D : E โ†’L[๐•œ] F)
(K : Submodule ๐•œ E) (hK : โˆ€ x โˆˆ K, T x โˆˆ K)
(hS : IsSelfAdjoint S) (hDS : โˆ€ x, S (D x) = D (T x))
(hcover : โˆ€ x : E, โˆƒ z โˆˆ K, D z = D x)
{ฯ : โ„} (hฯ : 0 โ‰ค ฯ) (hT : โ€–T.restrict hKโ€– โ‰ค ฯ) :
โ€–restrict T S D hDSโ€– โ‰ค ฯ := by
apply ContinuousLinearMap.opNorm_le_bound _ hฯ
intro y
exact norm_apply_le_of_source_subspace T S D K hK hS hDS hcover hฯ hT y.property
lemma norm_restrict_le_of_kills_direction
(T : E โ†’L[๐•œ] E) (S : F โ†’L[๐•œ] F) (D : E โ†’L[๐•œ] F)
(โ„“ : E โ†’L[๐•œ] ๐•œ) (e : E) (he : โ„“ e = 1) (hDe : D e = 0)
(hK : โˆ€ x โˆˆ LinearMap.ker (โ„“ : E โ†’โ‚—[๐•œ] ๐•œ),
T x โˆˆ LinearMap.ker (โ„“ : E โ†’โ‚—[๐•œ] ๐•œ))
(hS : IsSelfAdjoint S) (hDS : โˆ€ x, S (D x) = D (T x))
{ฯ : โ„} (hฯ : 0 โ‰ค ฯ) (hT : โ€–T.restrict hKโ€– โ‰ค ฯ) :
โ€–restrict T S D hDSโ€– โ‰ค ฯ :=
norm_restrict_le_of_source_subspace T S D _ hK hS hDS
(source_cover_of_kills_direction D โ„“ e he hDe) hฯ hT
end SpectralClosure
end Problem367
end
end OAI