Paper
Constructing the Oseledets decomposition with subspace growth estimates
18 pages · 26 marked passages · pdflatex · download PDF · lax-606786
-
Growth, Bernstein numbers and the compactness seminorm of an operator
Let be a bounded linear operator between real normed spaces.
- The growth of on a subspace is the slowest growth of a unit vector of ,with the convention (so ).
- For and , the -th Bernstein number isIn particular , and when .
- The compactness seminorm of is its distance to the compact operators,
- For , the -th Gelfand number of is
- The volume growth is the largest volume of the image of a unit parallelepiped:(Each infimum or supremum of an empty or unbounded set is taken to be .)
1 import Lax606786.BanachGrassmannian 2 import Mathlib.Analysis.Normed.Operator.Compact.Basic 3 … module docstring, 23 lines 27 28 namespace Lax606786.OperatorStatistics 29 30 open Lax606786.BanachGrassmannian 31 32 variable {X Y : Type*} [NormedAddCommGroup X] [NormedSpace ℝ X] 33 [NormedAddCommGroup Y] [NormedSpace ℝ Y] 34 35 /-- `g(T, V) = inf {‖T x‖ : x ∈ V, ‖x‖ = 1}`. -/ 36 noncomputable def growth (T : X →L[ℝ] Y) (V : Submodule ℝ X) : ℝ := 37 sInf ((fun x : X => ‖T x‖) '' (Metric.sphere (0 : X) 1 ∩ (V : Set X))) 38 39 /-- `ρ_k(T) = sup {g(T, V) : V ∈ 𝒢_k X}`. -/ 40 noncomputable def bernsteinNumber (T : X →L[ℝ] X) (k : ℕ) : ℝ := 41 sSup (Set.range fun V : GrassmannianFin X k => growth T (V.1 : Submodule ℝ X)) 42 43 /-- `‖T‖_c = inf {‖T - K‖ : K compact}`. -/ 44 noncomputable def compactSeminorm (T : X →L[ℝ] X) : ℝ := 45 sInf ((fun K : X →L[ℝ] X => ‖T - K‖) '' {K | IsCompactOperator K}) 46 47 /-- `s_k(T) = inf {‖T|_V‖ : V ∈ 𝒢^{k-1} X}`. -/ 48 noncomputable def gelfandNumber (T : X →L[ℝ] X) (k : ℕ) : ℝ := 49 sInf (Set.range fun V : GrassmannianCofin X (k - 1) => 50 ‖T.comp (V.1.1 : Submodule ℝ X).subtypeL‖) 51 52 open Classical in 53 /-- `D_k(T)`: the supremum over unit `k`-tuples `u` of `∏ᵢ d(T uᵢ, span {T u_j : j < i})`. -/ 54 noncomputable def volumeGrowth (T : X →L[ℝ] X) (k : ℕ) : ℝ := 55 sSup (Set.range fun u : Fin k → X => 56 if ∀ i, u i ∈ Metric.sphere (0 : X) 1 then 57 ∏ i : Fin k, Metric.infDist (T (u i)) 58 ((Submodule.span ℝ ((fun j : Fin k => T (u j)) '' {j | j < i})) : Set X) 59 else 0) 60 61 end Lax606786.OperatorStatistics 62 -
Cocycles and their Lyapunov exponents
A cocycle consists of an invertible ergodic measure-preserving transformation of a nonatomic standard probability space , together with a map from into the bounded operators on a real Banach space which is
- strongly measurable: is measurable for each ;
- forward integrable: .
No invertibility of is assumed. Its iterates are and .
All logarithms below take values in with . Exponents and multiplicities are indexed from , as usual; the value at index is not used.
- The growth rate of a vector is .
- The -th Lyapunov exponent () is , with the Bernstein number. The sequence is non-increasing and is the top exponent.
- The distinct exponents are the distinct values of : and for the least with . The multiplicity of is the number of with . When there is no exponent after the recursion stops: is recorded as , and the later 's are (the index at which they would start is recorded as , and ).
- The Bernstein index is . It stands in for the usual index of compactness, the growth rate of ; the concept shows that the two agree.
- The cocycle is quasicompact if almost everywhere.
- A family of operators is tempered if is tempered for .
1 import Lax606786.ExtendedLog 2 import Lax606786.OperatorStatistics 3 import Lax606786.TemperedFunctions 4 import Mathlib.Dynamics.Ergodic.Ergodic 5 import Mathlib.MeasureTheory.Function.L1Space.Integrable 6 … module docstring, 38 lines 45 46 namespace Lax606786.Cocycles 47 48 open MeasureTheory Filter 49 open Lax606786.OperatorStatistics Lax606786.ExtendedLog Lax606786.TemperedFunctions 50 51 variable {Ω X : Type*} [MeasurableSpace Ω] [StandardBorelSpace Ω] 52 [NormedAddCommGroup X] [NormedSpace ℝ X] [MeasurableSpace X] 53 54 /-- A strongly measurable, forward-integrable cocycle of bounded operators on `X` over an 55 invertible ergodic transformation of a Lebesgue probability space. -/ 56 structure Cocycle (Ω X : Type*) [MeasurableSpace Ω] [StandardBorelSpace Ω] 57 [NormedAddCommGroup X] [NormedSpace ℝ X] [MeasurableSpace X] where 58 /-- The base transformation, a measurable bijection with measurable inverse. -/ 59 σ : Ω ≃ᵐ Ω 60 /-- The invariant measure. -/ 61 μ : Measure Ω 62 /-- The generator `ω ↦ 𝓛_ω`. -/ 63 L : Ω → X →L[ℝ] X 64 isProbability : IsProbabilityMeasure μ 65 nullSingleton : NullSingletonClass μ 66 ergodic : Ergodic σ μ 67 stronglyMeasurable : ∀ x : X, Measurable (fun ω => L ω x) 68 forwardIntegrable : Integrable (fun ω => Real.log (max ‖L ω‖ 1)) μ 69 70 attribute [instance] Cocycle.isProbability Cocycle.nullSingleton 71 72 namespace Cocycle 73 74 /-- `𝓛^{(n)}_ω = 𝓛_{σ^{n-1}ω} ∘ ⋯ ∘ 𝓛_ω`. -/ 75 noncomputable def iterate (R : Cocycle Ω X) : ℕ → Ω → X →L[ℝ] X 76 | 0, _ => ContinuousLinearMap.id ℝ X 77 | n + 1, ω => (R.L (R.σ^[n] ω)).comp (R.iterate n ω) 78 79 /-- `λ_ω(x) = limsup (1/n) log ‖𝓛^{(n)}_ω x‖`. -/ 80 noncomputable def lambdaAt (R : Cocycle Ω X) (ω : Ω) (x : X) : EReal := 81 limsup (fun n : ℕ => logEReal ‖R.iterate n ω x‖ / (n : EReal)) atTop 82 83 /-- `χ_k(ω) = limsup (1/n) log ρ_k(𝓛^{(n)}_ω)`. Meaningful for `k ≥ 1`; `χ_0 = -∞`. -/ 84 noncomputable def chi (R : Cocycle Ω X) (k : ℕ) (ω : Ω) : EReal := 85 limsup (fun n : ℕ => logEReal (bernsteinNumber (R.iterate n ω) k) / (n : EReal)) atTop 86 87 /-- The index `k` at which the distinct exponent `λ_i` first occurs among the `χ_k`: `1` for 88 `i = 1`, then the least `t ≥ 1` with `χ_t` below `λ_{i-1}`, and `0` once there is none. Index `0` is 89 not used. -/ 90 noncomputable def lambdaIdx (R : Cocycle Ω X) : ℕ → Ω → ℕ 91 | 0, _ => 0 92 | 1, _ => 1 93 | (i + 2), ω => sInf {t : ℕ | 1 ≤ t ∧ R.chi t ω < R.chi (R.lambdaIdx (i + 1) ω) ω} 94 95 /-- The distinct exponent `λ_i` (for `i ≥ 1`). -/ 96 noncomputable def lambda (R : Cocycle Ω X) (i : ℕ) (ω : Ω) : EReal := 97 R.chi (R.lambdaIdx i ω) ω 98 99 /-- The multiplicity `m_i` of `λ_i` (for `i ≥ 1`); `0` if there is no exponent `λ_i`. -/ 100 noncomputable def mult (R : Cocycle Ω X) (i : ℕ) (ω : Ω) : ℕ := 101 R.lambdaIdx (i + 1) ω - R.lambdaIdx i ω 102 103 /-- The Bernstein index `ν(ω) = inf_{k ≥ 1} χ_k(ω)`. -/ 104 noncomputable def nu (R : Cocycle Ω X) (ω : Ω) : EReal := ⨅ k : ℕ, R.chi (k + 1) ω 105 106 /-- `ν < λ₁` almost everywhere. -/ 107 def IsQuasicompact (R : Cocycle Ω X) : Prop := 108 ∀ᵐ ω ∂(R.μ), R.nu ω < R.chi 1 ω 109 110 /-- `ω ↦ log ‖P_ω‖` is tempered. -/ 111 abbrev IsTempered (R : Cocycle Ω X) (P : Ω → X →L[ℝ] X) : Prop := 112 Tempered R.σ R.μ (fun ω => Real.log ‖P ω‖) 113 114 end Cocycle 115 116 end Lax606786.Cocycles 117 -
The semi-invertible Oseledets decomposition
Every cocycle on a separable Banach space , over an invertible ergodic base , has an Oseledets decomposition .
That is, the distinct Lyapunov exponents and their multiplicities are almost everywhere constant, and there are measurable equivariant fast spaces of dimension (), growing at rate both in norm and in slowest growth, together with strongly measurable tempered projections () onto invariant slow spaces , such that almost surely
The decomposition is nontrivial () exactly when is quasicompact.
By the concepts and , this happens if and only if the index of compactness , which always exists, is smaller than .
This is Theorem 3 of Lee (2024), a form of Oseledets' multiplicative ergodic theorem (Oseledets 1968; see also Raghunathan 1979, and Ruelle 1979 for compact operators on Hilbert space). Earlier Banach-space versions assumed the operators quasicompact (Thieullen 1987) or injective (Lian and Lu 2010). The semi-invertible form, in which the operators need not be invertible, is due to Froyland, Lloyd and Quas (2010) in finite dimensions and (2013) for quasicompact cocycles on Banach spaces, and to González-Tokman and Quas (2014) on separable Banach spaces.
1 import Lax606786.OseledetsDecompositions 2 … module docstring, 30 lines 33 34 namespace Lax606786.OseledetsDecomposition 35 36 open TopologicalSpace 37 open Lax606786.BanachGrassmannian Lax606786.Cocycles Lax606786.OseledetsDecompositions 38 39 /-- Every cocycle on a separable Banach space has an Oseledets decomposition. (On the zero 40 space it is the empty one: `L = 1`, `λ₁ = -∞`, no fast spaces.) -/ 41 axiom oseledets_decomposition {Ω X : Type*} [MeasurableSpace Ω] [StandardBorelSpace Ω] 42 [NormedAddCommGroup X] [NormedSpace ℝ X] [MeasurableSpace X] 43 (R : Cocycle Ω X) [SeparableSpace X] [BorelSpace X] [CompleteSpace X] : 44 ∃ (L : ℕ∞) (lam : ℕ → EReal) (m : ℕ → ℕ+) 45 (E : ∀ i : ℕ, 1 ≤ i → (i : ℕ∞) < L → Ω → GrassmannianFin X (m i)) 46 (V : ℕ → Ω → Submodule ℝ X) (P : ℕ → Ω → X →L[ℝ] X), 47 IsOseledetsDecomposition R L lam m E V P 48 49 end Lax606786.OseledetsDecomposition 50 -
The Grassmannians of a Banach space
Let be a real normed space. The unit sphere of a subspace is ; it spans , so it determines .
- The Grassmannian is the set of closed subspaces admitting a bounded projection . It carries the distancethe Hausdorff distance between unit spheres (infinite exactly when one of is and the other is not).
- For , is the set of of codimension , that is, with .
- For , is the set of -dimensional subspaces of . For the unit sphere is a nonempty compact set and carries the metric .
Each is equipped with the Borel -algebra of its distance.
1 import Mathlib.Analysis.RCLike.Lemmas 2 import Mathlib.MeasureTheory.Constructions.BorelSpace.Basic 3 import Mathlib.Topology.MetricSpace.Closeds 4 import Mathlib.Analysis.LocallyConvex.HahnBanach 5 import Mathlib.Analysis.Normed.Module.HahnBanach 6 … module docstring, 21 lines 28 29 namespace Lax606786.BanachGrassmannian 30 31 open TopologicalSpace 32 33 variable (X : Type*) [NormedAddCommGroup X] [NormedSpace ℝ X] 34 35 /-- `𝒢X`: the closed subspaces of `X` admitting a bounded projection onto them. -/ 36 def Grassmannian : Type _ := 37 {V : Submodule ℝ X // V.ClosedComplemented} 38 39 variable {X} 40 41 /-- Every subspace is spanned by its unit sphere. -/ 42 theorem span_sphere_inter (V : Submodule ℝ X) : 43 Submodule.span ℝ (Metric.sphere (0 : X) 1 ∩ (V : Set X)) = V := by 44 apply le_antisymm 45 · exact Submodule.span_le.2 Set.inter_subset_right 46 · intro x hx 47 rcases eq_or_ne x 0 with rfl | hx0 48 · exact Submodule.zero_mem _ 49 · have hxeq : x = ‖x‖ • (‖x‖⁻¹ • x) := by 50 rw [smul_smul, mul_inv_cancel₀ (norm_ne_zero_iff.2 hx0), one_smul] 51 rw [hxeq] 52 refine Submodule.smul_mem _ _ (Submodule.subset_span ⟨?_, V.smul_mem _ hx⟩) 53 rw [mem_sphere_zero_iff_norm, norm_smul, norm_inv, norm_norm, 54 inv_mul_cancel₀ (norm_ne_zero_iff.2 hx0)] 55 56 namespace Grassmannian 57 58 /-- The unit sphere `𝕊_V = {x ∈ V : ‖x‖ = 1}`. -/ 59 def sphere (V : Grassmannian X) : Set X := 60 Metric.sphere (0 : X) 1 ∩ (V.1 : Set X) 61 62 /-- `V ↦ 𝕊_V`, as a closed subset of `X`. -/ 63 def toCloseds (V : Grassmannian X) : Closeds X := 64 ⟨V.sphere, Metric.isClosed_sphere.inter V.2.isClosed⟩ 65 66 /-- A subspace is the span of its unit sphere, so `V ↦ 𝕊_V` is injective. -/ 67 theorem toCloseds_injective : Function.Injective (toCloseds (X := X)) := by 68 intro V W h 69 apply Subtype.ext 70 rw [← span_sphere_inter V.1, ← span_sphere_inter W.1] 71 exact congrArg (fun s : Closeds X => Submodule.span ℝ (s : Set X)) h 72 73 /-- The distance on `𝒢X`: the Hausdorff distance between unit spheres. -/ 74 noncomputable instance instEMetricSpace : EMetricSpace (Grassmannian X) := 75 EMetricSpace.induced toCloseds toCloseds_injective inferInstance 76 77 /-- The Borel `σ`-algebra of that distance. -/ 78 noncomputable instance instMeasurableSpace : MeasurableSpace (Grassmannian X) := 79 borel (Grassmannian X) 80 81 instance instBorelSpace : BorelSpace (Grassmannian X) := ⟨rfl⟩ 82 83 /-- `𝕊_V` is closed. -/ 84 theorem isClosed_sphere (V : Grassmannian X) : IsClosed V.sphere := 85 Metric.isClosed_sphere.inter V.2.isClosed 86 87 end Grassmannian 88 89 variable (X) 90 91 /-- `𝒢^k X`: the subspaces in `𝒢X` of codimension `k`. -/ 92 def GrassmannianCofin (k : ℕ) : Type _ := 93 {V : Grassmannian X // FiniteDimensional ℝ (X ⧸ (V.1 : Submodule ℝ X)) ∧ 94 Module.finrank ℝ (X ⧸ (V.1 : Submodule ℝ X)) = k} 95 96 /-- `𝒢_k X`: the `k`-dimensional subspaces of `X`. -/ 97 def GrassmannianFin (k : ℕ) : Type _ := 98 {V : Submodule ℝ X // FiniteDimensional ℝ V ∧ Module.finrank ℝ V = k} 99 100 variable {X} 101 102 namespace GrassmannianFin 103 104 /-- The unit sphere `𝕊_V = {x ∈ V : ‖x‖ = 1}`. -/ 105 def sphere {k : ℕ} (V : GrassmannianFin X k) : Set X := 106 Metric.sphere (0 : X) 1 ∩ (V.1 : Set X) 107 108 /-- `V` is the span of its unit sphere. -/ 109 theorem span_sphere {k : ℕ} (V : GrassmannianFin X k) : 110 Submodule.span ℝ V.sphere = (V.1 : Submodule ℝ X) := by 111 apply le_antisymm 112 · exact Submodule.span_le.2 Set.inter_subset_right 113 · intro x hx 114 rcases eq_or_ne x 0 with rfl | hx0 115 · exact Submodule.zero_mem _ 116 · have hxeq : x = ‖x‖ • (‖x‖⁻¹ • x) := by 117 rw [smul_smul, mul_inv_cancel₀ (norm_ne_zero_iff.2 hx0), one_smul] 118 rw [hxeq] 119 refine Submodule.smul_mem _ _ (Submodule.subset_span ⟨?_, V.1.smul_mem _ hx⟩) 120 rw [mem_sphere_zero_iff_norm, norm_smul, norm_inv, norm_norm, 121 inv_mul_cancel₀ (norm_ne_zero_iff.2 hx0)] 122 123 /-- `𝕊_V` is compact. -/ 124 theorem isCompact_sphere {k : ℕ} (V : GrassmannianFin X k) : IsCompact V.sphere := by 125 have : FiniteDimensional ℝ (V.1 : Submodule ℝ X) := V.2.1 126 have himg : V.sphere = 127 (Subtype.val : (V.1 : Submodule ℝ X) → X) '' Metric.sphere (0 : (V.1 : Submodule ℝ X)) 1 := by 128 ext x 129 simp only [sphere, Set.mem_inter_iff, Set.mem_image] 130 constructor 131 · rintro ⟨hx1, hxV⟩ 132 refine ⟨⟨x, hxV⟩, ?_, rfl⟩ 133 rwa [mem_sphere_zero_iff_norm] at hx1 ⊢ 134 · rintro ⟨v, hv, rfl⟩ 135 rw [mem_sphere_zero_iff_norm] at hv 136 exact ⟨by rwa [mem_sphere_zero_iff_norm], v.2⟩ 137 rw [himg] 138 exact (_root_.isCompact_sphere (0 : (V.1 : Submodule ℝ X)) 1).image continuous_subtype_val 139 140 /-- `𝕊_V` is nonempty when `k ≠ 0`. -/ 141 theorem nonempty_sphere {k : ℕ} [NeZero k] (V : GrassmannianFin X k) : V.sphere.Nonempty := by 142 have hne : (V.1 : Submodule ℝ X) ≠ ⊥ := by 143 intro h 144 apply NeZero.ne k 145 have hfr := V.2.2 146 rw [h, finrank_bot] at hfr 147 exact hfr.symm 148 obtain ⟨x, hxV, hx0⟩ := Submodule.exists_mem_ne_zero_of_ne_bot hne 149 refine ⟨‖x‖⁻¹ • x, ?_, V.1.smul_mem _ hxV⟩ 150 rw [mem_sphere_zero_iff_norm, norm_smul, norm_inv, norm_norm, 151 inv_mul_cancel₀ (norm_ne_zero_iff.2 hx0)] 152 153 /-- `V ↦ 𝕊_V`, as a nonempty compact subset of `X`. -/ 154 noncomputable def toNonemptyCompacts {k : ℕ} [NeZero k] (V : GrassmannianFin X k) : 155 NonemptyCompacts X where 156 carrier := V.sphere 157 isCompact' := V.isCompact_sphere 158 nonempty' := V.nonempty_sphere 159 160 /-- A subspace is the span of its unit sphere, so `V ↦ 𝕊_V` is injective. -/ 161 theorem toNonemptyCompacts_injective (k : ℕ) [NeZero k] : 162 Function.Injective (toNonemptyCompacts (X := X) (k := k)) := by 163 intro V W h 164 apply Subtype.ext 165 rw [← V.span_sphere, ← W.span_sphere] 166 exact congrArg (fun s : NonemptyCompacts X => Submodule.span ℝ (s : Set X)) h 167 168 /-- The metric on `𝒢_k X`: the Hausdorff distance between unit spheres. -/ 169 noncomputable instance instMetricSpace (k : ℕ) [NeZero k] : MetricSpace (GrassmannianFin X k) := 170 MetricSpace.induced toNonemptyCompacts (toNonemptyCompacts_injective k) inferInstance 171 172 /-- The Borel `σ`-algebra of that metric. -/ 173 noncomputable instance instMeasurableSpace (k : ℕ) [NeZero k] : 174 MeasurableSpace (GrassmannianFin X k) := 175 borel (GrassmannianFin X k) 176 177 instance instBorelSpace (k : ℕ) [NeZero k] : BorelSpace (GrassmannianFin X k) := ⟨rfl⟩ 178 179 /-- `V` is recovered from `𝕊_V`. -/ 180 theorem sphere_injective (k : ℕ) : Function.Injective (sphere (X := X) (k := k)) := by 181 intro V W h 182 apply Subtype.ext 183 rw [← V.span_sphere, ← W.span_sphere, h] 184 185 /-- A finite-dimensional subspace admits a bounded projection onto it, so `𝒢_k X` sits inside 186 `𝒢 X`. -/ 187 theorem closedComplemented {k : ℕ} (V : GrassmannianFin X k) : 188 (V.1 : Submodule ℝ X).ClosedComplemented := by 189 have : FiniteDimensional ℝ (V.1 : Submodule ℝ X) := V.2.1 190 exact Submodule.ClosedComplemented.of_finiteDimensional _ 191 192 /-- The inclusion `𝒢_k X → 𝒢 X`. -/ 193 def toGrassmannian {k : ℕ} (V : GrassmannianFin X k) : Grassmannian X := 194 ⟨V.1, V.closedComplemented⟩ 195 196 end GrassmannianFin 197 198 end Lax606786.BanachGrassmannian 199 -
The Grassmannian of k-dimensional subspaces is Polish
Let be a real normed space and , and give the Hausdorff distance between unit spheres. If is complete then so is , and if is separable then so is . In particular is Polish when is a separable Banach space. Completeness of spaces of closed subspaces under the gap distance is classical (Kato 2013, Chapter IV, §2.1).
1 import Lax606786.BanachGrassmannian 2 … module docstring, 11 lines 14 15 namespace Lax606786.GrassmannianPolish 16 17 open TopologicalSpace 18 open Lax606786.BanachGrassmannian 19 20 /-- `𝒢_k X` is complete when `X` is. -/ 21 axiom completeSpace_grassmannianFin {X : Type*} [NormedAddCommGroup X] [NormedSpace ℝ X] 22 [CompleteSpace X] (k : ℕ) [NeZero k] : CompleteSpace (GrassmannianFin X k) 23 24 /-- `𝒢_k X` is separable when `X` is. -/ 25 axiom separableSpace_grassmannianFin {X : Type*} [NormedAddCommGroup X] [NormedSpace ℝ X] 26 [SeparableSpace X] (k : ℕ) [NeZero k] : SeparableSpace (GrassmannianFin X k) 27 28 end Lax606786.GrassmannianPolish 29 -
Measurability of Bernstein numbers and of pushforwards
Let be a separable real normed space with a -algebra containing the open sets, and let be a strongly measurable family of bounded operators on (each measurable). Then for :
- is measurable, and so is ;
- is measurable for every measurable (if the -algebra of is Borel);
- is measurable for every subspace , and is measurable for every measurable (if the -algebra of is Borel);
- if moreover is complete and is measurable, then any with for every is measurable.
For the norm topology on the operators, is itself Borel measurable.
- thm✓
Lax606786.MeasurableOperatorFamilies(1st statement) - thm✓
Lax606786.MeasurableOperatorFamilies(2nd statement) - thm✓
Lax606786.MeasurableOperatorFamilies(3rd statement) - thm✓
Lax606786.MeasurableOperatorFamilies(4th statement) - thm✓
Lax606786.MeasurableOperatorFamilies(5th statement) - thm✓
Lax606786.MeasurableOperatorFamilies(6th statement) - thm✓
Lax606786.MeasurableOperatorFamilies(7th statement)
1 import Lax606786.OperatorStatistics 2 import Mathlib.MeasureTheory.Constructions.BorelSpace.Basic 3 4 import Mathlib.MeasureTheory.Constructions.BorelSpace.ContinuousLinearMap … module docstring, 21 lines 26 27 namespace Lax606786.MeasurableOperatorFamilies 28 29 open TopologicalSpace 30 open Lax606786.BanachGrassmannian Lax606786.OperatorStatistics 31 32 /-- `ω ↦ ‖T_ω‖` is measurable for a strongly measurable family `T` on a separable space. -/ 33 axiom measurable_opNorm {Ω X Y : Type*} [MeasurableSpace Ω] [NormedAddCommGroup X] 34 [NormedSpace ℝ X] [NormedAddCommGroup Y] [NormedSpace ℝ Y] [MeasurableSpace Y] [BorelSpace Y] 35 [SeparableSpace X] {T : Ω → X →L[ℝ] Y} (hT : ∀ x, Measurable fun ω => T ω x) : 36 Measurable fun ω => ‖T ω‖ 37 38 variable {X : Type*} [NormedAddCommGroup X] [NormedSpace ℝ X] 39 40 /-- `ω ↦ ρ_k(A_ω)` is measurable for strongly measurable `A`. -/ 41 axiom measurable_bernsteinNumber {Ω : Type*} [MeasurableSpace Ω] [MeasurableSpace X] 42 [OpensMeasurableSpace X] [SeparableSpace X] (k : ℕ) [NeZero k] {A : Ω → X →L[ℝ] X} 43 (hA : ∀ x : X, Measurable fun ω => A ω x) : 44 Measurable fun ω => bernsteinNumber (A ω) k 45 46 /-- A strongly measurable family applied to a measurable vector field is measurable. -/ 47 axiom measurable_apply {Ω : Type*} [MeasurableSpace Ω] [MeasurableSpace X] 48 [BorelSpace X] [SeparableSpace X] {A : Ω → X →L[ℝ] X} 49 (hA : ∀ x : X, Measurable fun ω => A ω x) {v : Ω → X} (hv : Measurable v) : 50 Measurable fun ω => A ω (v ω) 51 52 /-- `ω ↦ g(A_ω, V)` is measurable for a fixed subspace `V ≠ 0`. -/ 53 axiom measurable_growth {Ω : Type*} [MeasurableSpace Ω] [MeasurableSpace X] 54 [OpensMeasurableSpace X] [SeparableSpace X] {A : Ω → X →L[ℝ] X} 55 (hA : ∀ x : X, Measurable fun ω => A ω x) (V : Submodule ℝ X) 56 (hV : (Metric.sphere (0 : X) 1 ∩ (V : Set X)).Nonempty) : 57 Measurable fun ω => growth (A ω) V 58 59 /-- `ω ↦ g(A_ω, E(ω))` is measurable for a measurable family `E` in `𝒢_k X`. -/ 60 axiom measurable_growth_moving {Ω : Type*} [MeasurableSpace Ω] [MeasurableSpace X] 61 [BorelSpace X] [SeparableSpace X] {m : ℕ} [NeZero m] 62 {A : Ω → X →L[ℝ] X} (hA : ∀ x : X, Measurable fun ω => A ω x) 63 {E : Ω → GrassmannianFin X m} (hE : Measurable E) : 64 Measurable fun ω => growth (A ω) ((E ω).1 : Submodule ℝ X) 65 66 /-- `ω ↦ A_ω E(ω)` is measurable in `𝒢_k X`. -/ 67 axiom measurable_pushforward {Ω : Type*} [MeasurableSpace Ω] [MeasurableSpace X] 68 [OpensMeasurableSpace X] [SeparableSpace X] [CompleteSpace X] (k : ℕ) 69 [NeZero k] (E : Ω → GrassmannianFin X k) (hE : Measurable E) 70 (A : Ω → X →L[ℝ] X) (hA : ∀ x : X, Measurable fun ω => A ω x) 71 (F : Ω → GrassmannianFin X k) 72 (hF : ∀ ω, (F ω).1 = Submodule.map ((A ω : X →ₗ[ℝ] X)) (E ω).1) : 73 Measurable F 74 75 /-- For the norm topology on the operators, `T ↦ ρ_k(T)` is measurable. -/ 76 axiom measurable_bernsteinNumber_norm [SeparableSpace X] (k : ℕ) [NeZero k] 77 [Nonempty (GrassmannianFin X k)] : 78 Measurable (fun T : X →L[ℝ] X => bernsteinNumber T k) 79 80 end Lax606786.MeasurableOperatorFamilies 81 -
Elementary properties of the growth function
Let be real normed spaces, and bounded operators and a subspace. Then
- , and for every ;
- whenever ;
- for every unit vector (for );
- (the growth inequality)
- thm✓
Lax606786.GrowthFunctionProperties(1st statement) - thm✓
Lax606786.GrowthFunctionProperties(2nd statement) - thm✓
Lax606786.GrowthFunctionProperties(3rd statement) - thm✓
Lax606786.GrowthFunctionProperties(4th statement) - thm✓
Lax606786.GrowthFunctionProperties(5th statement) - thm✓
Lax606786.GrowthFunctionProperties(6th statement)
1 import Lax606786.OperatorStatistics 2 … module docstring, 14 lines 17 18 namespace Lax606786.GrowthFunctionProperties 19 20 open Lax606786.OperatorStatistics 21 22 variable {X Y Z : Type*} [NormedAddCommGroup X] [NormedSpace ℝ X] 23 [NormedAddCommGroup Y] [NormedSpace ℝ Y] [NormedAddCommGroup Z] [NormedSpace ℝ Z] 24 25 /-- `g(T, V) ≥ 0`. -/ 26 axiom growth_nonneg (T : X →L[ℝ] Y) (V : Submodule ℝ X) : 0 ≤ growth T V 27 28 /-- `g(T, V) ≤ ‖T‖`. -/ 29 axiom growth_le_opNorm (T : X →L[ℝ] Y) (V : Submodule ℝ X) : growth T V ≤ ‖T‖ 30 31 /-- `g(T, V) ‖u‖ ≤ ‖T u‖` for `u ∈ V`. -/ 32 axiom growth_mul_norm_le (T : X →L[ℝ] Y) {V : Submodule ℝ X} {u : X} (hu : u ∈ V) : 33 growth T V * ‖u‖ ≤ ‖T u‖ 34 35 /-- Shrinking the subspace can only increase the slowest growth. -/ 36 axiom growth_le_growth {T : X →L[ℝ] X} {W V : Submodule ℝ X} (hWV : W ≤ V) (hW : W ≠ ⊥) : 37 growth T V ≤ growth T W 38 39 /-- The growth on a line is the growth of its unit vector. -/ 40 axiom growth_span_singleton (T : X →L[ℝ] X) {u : X} (hu : ‖u‖ = 1) : 41 growth T (Submodule.span ℝ {u}) = ‖T u‖ 42 43 /-- The growth inequality for a composition. -/ 44 axiom growth_comp_bounds (T : X →L[ℝ] Y) (S : Y →L[ℝ] Z) (V : Submodule ℝ X) : 45 growth T V * growth S (V.map (T : X →ₗ[ℝ] Y)) ≤ growth (S.comp T) V ∧ 46 growth (S.comp T) V ≤ min (‖T‖ * growth S (V.map (T : X →ₗ[ℝ] Y))) (growth T V * ‖S‖) 47 48 end Lax606786.GrowthFunctionProperties 49 -
Volume growth is comparable to products of Bernstein numbers
Let be a real normed space and . There are constants , depending only on , such that for every bounded operator on
Moreover , and there is depending only on with
This is Lemma 8 of Lee (2024); the comparison follows the volume-growth arguments of González-Tokman and Quas (2015). Volume growth is also the basis of Blumenthal's (2016) approach to the multiplicative ergodic theorem.
- thm✓
Lax606786.VolumeGrowthComparison(1st statement) - thm✓
Lax606786.VolumeGrowthComparison(2nd statement) - thm✓
Lax606786.VolumeGrowthComparison(3rd statement)
1 import Lax606786.OperatorStatistics 2 … module docstring, 14 lines 17 18 namespace Lax606786.VolumeGrowthComparison 19 20 open Lax606786.OperatorStatistics 21 22 variable {X : Type*} [NormedAddCommGroup X] [NormedSpace ℝ X] 23 24 /-- `D_k` is comparable to `ρ_1 ⋯ ρ_k`, uniformly in the operator. -/ 25 axiom volumeGrowth_comparable (k : ℕ) : 26 ∃ c C : ℝ, 0 < c ∧ c ≤ C ∧ ∀ T : X →L[ℝ] X, 27 c * volumeGrowth T k ≤ ∏ i ∈ Finset.range k, bernsteinNumber T (i + 1) ∧ 28 ∏ i ∈ Finset.range k, bernsteinNumber T (i + 1) ≤ C * volumeGrowth T k 29 30 /-- `D_k` is submultiplicative. -/ 31 axiom volumeGrowth_submultiplicative (k : ℕ) (T S : X →L[ℝ] X) : 32 volumeGrowth (S.comp T) k ≤ volumeGrowth S k * volumeGrowth T k 33 34 /-- `ρ_1 ⋯ ρ_k` is submultiplicative up to a constant. -/ 35 axiom prod_bernsteinNumber_submultiplicative (k : ℕ) : 36 ∃ K : ℝ, 0 < K ∧ ∀ S T : X →L[ℝ] X, 37 ∏ i ∈ Finset.range k, bernsteinNumber (S.comp T) (i + 1) 38 ≤ K * ((∏ i ∈ Finset.range k, bernsteinNumber S (i + 1)) * 39 (∏ i ∈ Finset.range k, bernsteinNumber T (i + 1))) 40 41 end Lax606786.VolumeGrowthComparison 42 - thm✓
-
A measurable selection principle
Let be a sequence in a measurable space and a cover of by measurable sets. Then the first-index map is measurable, and so is .
A map into a separable metric space is measurable as soon as its distance to every point is.
1 import Mathlib.MeasureTheory.MeasurableSpace.Constructions 2 3 import Mathlib.MeasureTheory.Constructions.BorelSpace.Metric … module docstring, 11 lines 15 16 namespace Lax606786.MeasurableSelection 17 18 /-- The first index of a countable measurable cover is measurable. -/ 19 axiom measurable_of_countable_cover {Ω Y : Type*} [MeasurableSpace Ω] [MeasurableSpace Y] 20 (x : ℕ → Y) (S : ℕ → Set Ω) (hS : ∀ n, MeasurableSet (S n)) (hcover : (⋃ n, S n) = Set.univ) : 21 ∃ N : Ω → ℕ, Measurable N ∧ Measurable (fun ω => x (N ω)) ∧ 22 ∀ ω, ω ∈ S (N ω) ∧ ∀ m, ω ∈ S m → N ω ≤ m 23 24 /-- A map into a separable metric space is measurable when its distance to every point is. -/ 25 axiom measurable_of_measurable_dist {Y Ω : Type*} [MetricSpace Y] 26 [TopologicalSpace.SeparableSpace Y] [MeasurableSpace Y] [BorelSpace Y] [MeasurableSpace Ω] {F : Ω → Y} 27 (h : ∀ y : Y, Measurable fun ω => dist (F ω) y) : Measurable F 28 29 end Lax606786.MeasurableSelection 30 -
Elementary properties of Bernstein numbers
Let be a real normed space and bounded operators on .
- , , , and for .
- for every , and .
For every ,
Moreover, for the growth is jointly continuous in , for the operator norm on and the Hausdorff distance between unit spheres on .
- thm✓
Lax606786.BernsteinNumberProperties(1st statement) - thm✓
Lax606786.BernsteinNumberProperties(2nd statement) - thm✓
Lax606786.BernsteinNumberProperties(3rd statement) - thm✓
Lax606786.BernsteinNumberProperties(4th statement) - thm✓
Lax606786.BernsteinNumberProperties(5th statement) - thm✓
Lax606786.BernsteinNumberProperties(6th statement) - thm✓
Lax606786.BernsteinNumberProperties(7th statement) - thm✓
Lax606786.BernsteinNumberProperties(8th statement) - thm✓
Lax606786.BernsteinNumberProperties(9th statement) - thm✓
Lax606786.BernsteinNumberProperties(10th statement)
1 import Lax606786.OperatorStatistics 2 … module docstring, 17 lines 20 21 namespace Lax606786.BernsteinNumberProperties 22 23 open Lax606786.BanachGrassmannian Lax606786.OperatorStatistics 24 25 variable {X : Type*} [NormedAddCommGroup X] [NormedSpace ℝ X] 26 27 /-- `ρ_k(T) ≥ 0`. -/ 28 axiom bernsteinNumber_nonneg (T : X →L[ℝ] X) (k : ℕ) : 0 ≤ bernsteinNumber T k 29 30 /-- `ρ_0(T) = 0`. -/ 31 axiom bernsteinNumber_zero (T : X →L[ℝ] X) : bernsteinNumber T 0 = 0 32 33 /-- `ρ_1(T) = ‖T‖`. -/ 34 axiom bernsteinNumber_one (T : X →L[ℝ] X) : bernsteinNumber T 1 = ‖T‖ 35 36 /-- `ρ_k(T) ≤ ‖T‖`. -/ 37 axiom bernsteinNumber_le_opNorm (T : X →L[ℝ] X) (k : ℕ) : bernsteinNumber T k ≤ ‖T‖ 38 39 /-- `ρ_q(T) ≤ ρ_p(T)` for `1 ≤ p ≤ q`. -/ 40 axiom bernsteinNumber_antitone (T : X →L[ℝ] X) {p q : ℕ} (hp : 1 ≤ p) (hpq : p ≤ q) : 41 bernsteinNumber T q ≤ bernsteinNumber T p 42 43 /-- `g(T, V) ≤ ρ_m(T)` for `V ∈ 𝒢_m X`. -/ 44 axiom growth_le_bernsteinNumber {m : ℕ} (T : X →L[ℝ] X) (V : GrassmannianFin X m) : 45 growth T (V.1 : Submodule ℝ X) ≤ bernsteinNumber T m 46 47 /-- `ρ_1(T) ⋯ ρ_k(T) ≤ ‖T‖^k`. -/ 48 axiom prod_bernsteinNumber_le_opNorm_pow (T : X →L[ℝ] X) (k : ℕ) : 49 ∏ i ∈ Finset.range k, bernsteinNumber T (i + 1) ≤ ‖T‖ ^ k 50 51 /-- `ρ_k(S ∘ T) ≤ ρ_k(S) ‖T‖`. -/ 52 axiom bernsteinNumber_comp_le_mul_norm (S T : X →L[ℝ] X) (k : ℕ) : 53 bernsteinNumber (S.comp T) k ≤ bernsteinNumber S k * ‖T‖ 54 55 /-- `ρ_k(S ∘ T) ≤ ‖S‖ ρ_k(T)`. -/ 56 axiom bernsteinNumber_comp_le_norm_mul (S T : X →L[ℝ] X) (k : ℕ) : 57 bernsteinNumber (S.comp T) k ≤ ‖S‖ * bernsteinNumber T k 58 59 /-- `(T, V) ↦ g(T, V)` is continuous on `𝓑(X) × 𝒢_k X`. -/ 60 axiom continuous_growth (k : ℕ) [NeZero k] : 61 Continuous (fun p : (X →L[ℝ] X) × GrassmannianFin X k => growth p.1 (p.2.1 : Submodule ℝ X)) 62 63 end Lax606786.BernsteinNumberProperties 64 -
The Grassmannian contraction estimate
Let be a real normed space, , a bounded operator on and with . Let with and . Then and, writing ,
So an operator with a gap between and contracts the subspaces on which it grows fast (Lee 2024, Lemma 12).
1 import Lax606786.OperatorStatistics 2 … module docstring, 13 lines 16 17 namespace Lax606786.GrassmannianContraction 18 19 open Lax606786.BanachGrassmannian Lax606786.OperatorStatistics 20 21 /-- The images of two fast-growing `k`-dimensional subspaces are close. -/ 22 axiom dist_image_le {X : Type*} [NormedAddCommGroup X] [NormedSpace ℝ X] 23 (k : ℕ) [NeZero k] (T : X →L[ℝ] X) (Θ : ℝ) 24 (hΘ : bernsteinNumber T (k + 1) < Θ) (hΘ' : Θ < bernsteinNumber T k) 25 (V W : GrassmannianFin X k) 26 (hV : Θ < growth T (V.1 : Submodule ℝ X)) (hW : Θ < growth T (W.1 : Submodule ℝ X)) 27 (TV TW : GrassmannianFin X k) 28 (hTV : (TV.1 : Submodule ℝ X) = (V.1 : Submodule ℝ X).map (T : X →ₗ[ℝ] X)) 29 (hTW : (TW.1 : Submodule ℝ X) = (W.1 : Submodule ℝ X).map (T : X →ₗ[ℝ] X)) : 30 dist TV TW ≤ 4 * (bernsteinNumber T (k + 1) / Θ) / (1 - bernsteinNumber T (k + 1) / Θ) 31 32 end Lax606786.GrassmannianContraction 33 -
Measurable bases and measurable splittings of subspaces
Let be a separable Banach space with its Borel -algebra and .
- There is a measurable map such that spans for every . Composing with a measurable family gives a measurable choice of basis of .
- There are measurable maps and such that is a unit vector and for every .
Each -dimensional subspace also contains a -dimensional subspace chosen measurably.
- thm✓
Lax606786.MeasurableBases(1st statement) - thm✓
Lax606786.MeasurableBases(2nd statement) - thm✓
Lax606786.MeasurableBases(3rd statement)
1 import Lax606786.BanachGrassmannian 2 import Mathlib.MeasureTheory.Constructions.BorelSpace.Basic 3 4 import Mathlib.Analysis.InnerProductSpace.PiL2 … module docstring, 16 lines 21 22 namespace Lax606786.MeasurableBases 23 24 open TopologicalSpace 25 open Lax606786.BanachGrassmannian 26 27 variable {X : Type*} [NormedAddCommGroup X] [NormedSpace ℝ X] 28 29 /-- A measurable spanning `k`-tuple for each `V ∈ 𝒢_k X`. -/ 30 axiom exists_measurable_basis [SeparableSpace X] [CompleteSpace X] [MeasurableSpace X] 31 [BorelSpace X] (k : ℕ) [NeZero k] : 32 ∃ b : GrassmannianFin X k → Fin k → X, 33 Measurable b ∧ ∀ V, (V.1 : Submodule ℝ X) = Submodule.span ℝ (Set.range (b V)) 34 35 /-- A measurable splitting of a `(k+1)`-dimensional subspace into a `k`-dimensional one and a 36 unit vector. -/ 37 axiom exists_measurable_retract [SeparableSpace X] [CompleteSpace X] [MeasurableSpace X] 38 [BorelSpace X] (k : ℕ) [NeZero k] : 39 ∃ (U : GrassmannianFin X (k + 1) → GrassmannianFin X k) (u : GrassmannianFin X (k + 1) → X), 40 Measurable U ∧ Measurable u ∧ 41 ∀ V, ‖u V‖ = 1 ∧ (U V).1 ⊔ Submodule.span ℝ {u V} = V.1 42 43 /-- A measurable choice of a codimension-one subspace of each `(k+1)`-dimensional subspace. -/ 44 axiom exists_measurable_hyperplane [SeparableSpace X] [CompleteSpace X] [MeasurableSpace X] 45 [BorelSpace X] (k : ℕ) [NeZero k] : 46 ∃ U : GrassmannianFin X (k + 1) → GrassmannianFin X k, Measurable U ∧ 47 ∀ V, (U V).1 < V.1 48 49 end Lax606786.MeasurableBases 50 -
Bernstein numbers under a projection
Let be a real normed space and a bounded operator on .
- (Small growth.) If , is a bounded projection onto and , then .
- (Large growth.) If , are bounded projections onto , , , and , then for every ,
These are Lemmas 16 and 17 of Lee (2024).
- thm✓
Lax606786.GrowthUnderProjections(1st statement) - thm✓
Lax606786.GrowthUnderProjections(2nd statement)
1 import Lax606786.OperatorStatistics 2 … module docstring, 15 lines 18 19 namespace Lax606786.GrowthUnderProjections 20 21 open Lax606786.BanachGrassmannian Lax606786.OperatorStatistics 22 23 variable {X : Type*} [NormedAddCommGroup X] [NormedSpace ℝ X] 24 25 /-- Compressing to a codimension-`k` subspace costs at most `k` Bernstein indices. -/ 26 axiom bernsteinNumber_add_le_comp_proj (k l : ℕ) (hl : l ≠ 0) (V : GrassmannianCofin X k) 27 (P : X →L[ℝ] X) (hP : ∀ x, P x ∈ (V.1.1 : Submodule ℝ X)) 28 (hidem : ∀ x ∈ (V.1.1 : Submodule ℝ X), P x = x) (T : X →L[ℝ] X) : 29 bernsteinNumber T (k + l) ≤ bernsteinNumber (T.comp P) l 30 31 /-- When `T` grows faster on `ker P` than on `range P`, compressing to `range P` loses at most 32 `k` Bernstein indices, up to the factor `4 ‖P‖ ‖P'‖`. -/ 33 axiom bernsteinNumber_comp_proj_le (k l : ℕ) (V V' : GrassmannianCofin X k) (E : Submodule ℝ X) 34 (P P' T : X →L[ℝ] X) 35 (hPproj : LinearMap.IsProj (V.1.1 : Submodule ℝ X) (P : X →ₗ[ℝ] X)) 36 (hPker : LinearMap.ker (P : X →ₗ[ℝ] X) = E) 37 (hP'proj : LinearMap.IsProj (V'.1.1 : Submodule ℝ X) (P' : X →ₗ[ℝ] X)) 38 (hPcompat : P'.comp T = T.comp P) 39 (hfast : growth T E > ‖T.comp (V.1.1 : Submodule ℝ X).subtypeL‖) : 40 bernsteinNumber (T.comp P) l ≤ 4 * ‖P‖ * ‖P'‖ * bernsteinNumber T (l + k) 41 42 end Lax606786.GrowthUnderProjections 43 -
Subadditive families of measurable functions
Let be a map on a measurable space. A sequence of measurable functions is a subadditive family over if
The typical example is for a cocycle of bounded operators. Compare the subadditive processes of Krengel (1985).
1 import Mathlib.MeasureTheory.Constructions.BorelSpace.Real 2 … module docstring, 13 lines 16 17 namespace Lax606786.SubadditiveFamilies 18 19 /-- `(f_n)` is a subadditive family of measurable functions over `σ`. -/ 20 def IsSubadditiveFamily {Ω : Type*} [MeasurableSpace Ω] (σ : Ω → Ω) (f : ℕ → Ω → ℝ) : Prop := 21 (∀ n, Measurable (f n)) ∧ ∀ m n ω, f (m + n) ω ≤ f m (σ^[n] ω) + f n ω 22 23 end Lax606786.SubadditiveFamilies 24 -
Kingman's subadditive ergodic theorem
Let be an ergodic measure-preserving transformation of a probability space and a subadditive family over with integrable. Let denote the lower integral. Then there is a constant with
such that for -a.e. . Limits are taken in the extended reals. When the are integrable, .
The theorem is due to Kingman (1968); the proof follows Steele (1989).
1 import Lax606786.SubadditiveFamilies 2 import Lax606786.LowerIntegrals 3 import Mathlib.Dynamics.Ergodic.Ergodic 4 import Mathlib.MeasureTheory.Function.L1Space.Integrable 5 … module docstring, 14 lines 20 21 namespace Lax606786.KingmanTheorem 22 23 open MeasureTheory Filter Topology 24 open Lax606786.SubadditiveFamilies Lax606786.LowerIntegrals 25 26 variable {Ω : Type*} [MeasurableSpace Ω] 27 28 /-- Kingman's theorem: `f_n / n` converges almost everywhere to the constant 29 `C = lim L_μ(f_n / n) = inf L_μ(f_n / n) ∈ [-∞, ∞)`. -/ 30 axiom kingman (μ : Measure Ω) [IsProbabilityMeasure μ] (σ : Ω → Ω) (hσ : Ergodic σ μ) 31 (f : ℕ → Ω → ℝ) (hf : IsSubadditiveFamily σ f) 32 (h1 : Integrable (fun ω => max (f 1 ω) 0) μ) : 33 ∃ C : EReal, C < ⊤ ∧ 34 (∀ᵐ ω ∂μ, Tendsto (fun n => (f n ω / n : EReal)) atTop (𝓝 C)) ∧ 35 (∀ n : ℕ, 1 ≤ n → C ≤ lowerIntegral μ (fun ω => f n ω / n)) ∧ 36 Tendsto (fun n : ℕ => lowerIntegral μ (fun ω => f n ω / n)) atTop (𝓝 C) 37 38 end Lax606786.KingmanTheorem 39 -
The backward Vitali lemma
Let be an invertible ergodic measure-preserving transformation of a nonatomic standard probability space . For and write for the orbit window of radius .
Let , and let be measurable with for every . Then there are a measurable set and a measurable such that
- for every , for some ;
- the windows , , are pairwise disjoint;
- .
This is a form of Rudolph's backward Vitali lemma (Rudolph 1990, Theorem 3.9).
1 import Mathlib.Dynamics.Ergodic.Ergodic 2 import Mathlib.MeasureTheory.Measure.Typeclasses.NullSingletonClass 3 import Mathlib.MeasureTheory.Constructions.Polish.Basic 4 … module docstring, 19 lines 24 25 namespace Lax606786.BackwardVitali 26 27 open MeasureTheory Filter 28 29 /-- The orbit window `{σⁱ ω : -j ≤ i ≤ j}`. -/ 30 def orbitWindow {Ω : Type*} [MeasurableSpace Ω] (σ : Ω ≃ᵐ Ω) (ω : Ω) (j : ℕ) : Set Ω := 31 {x | ∃ i ≤ j, x = σ^[i] ω ∨ x = (σ.symm : Ω → Ω)^[i] ω} 32 33 /-- Disjoint orbit windows, of admissible radii, covering all but `ε` of the space. -/ 34 axiom backwardVitali {Ω : Type*} [MeasurableSpace Ω] [StandardBorelSpace Ω] (μ : Measure Ω) 35 [IsProbabilityMeasure μ] [NullSingletonClass μ] (σ : Ω ≃ᵐ Ω) (hσ : Ergodic σ μ) 36 (ε : ℝ) (hε : 0 < ε) (j : ℕ → Ω → ℕ) (hj : ∀ k, Measurable (j k)) 37 (hjtop : ∀ ω, Tendsto (fun k => j k ω) atTop atTop) : 38 ∃ (A : Set Ω) (jω : Ω → ℕ), MeasurableSet A ∧ Measurable jω ∧ 39 (∀ ω ∈ A, ∃ k, jω ω = j k ω) ∧ 40 (A.PairwiseDisjoint fun ω => orbitWindow σ ω (jω ω)) ∧ 41 ENNReal.ofReal (1 - ε) < μ (⋃ ω ∈ A, orbitWindow σ ω (jω ω)) 42 43 end Lax606786.BackwardVitali 44 -
Kingman's theorem for balanced intervals
Let be an invertible ergodic measure-preserving transformation of a nonatomic standard probability space , and let be a subadditive family over with integrable. Then the averages over the balanced time intervals converge to the same constant as in Kingman's theorem:
with , limits taken in the extended reals, and the lower integral. This is Theorem 24 of Lee (2024).
Given that the forward averages converge almost everywhere to a constant , the shifted averages and the balanced averages converge to the same .
1 import Lax606786.SubadditiveFamilies 2 import Lax606786.LowerIntegrals 3 import Mathlib.Dynamics.Ergodic.Ergodic 4 import Mathlib.MeasureTheory.Integral.Bochner.Basic 5 import Mathlib.MeasureTheory.Measure.Typeclasses.NullSingletonClass 6 import Mathlib.MeasureTheory.Constructions.Polish.Basic 7 import Mathlib.Data.EReal.Operations 8 … module docstring, 18 lines 27 28 namespace Lax606786.BalancedKingman 29 30 open MeasureTheory Filter Topology 31 open Lax606786.SubadditiveFamilies 32 33 /-- `f_{2n}(σ^{-n} ω) / 2n` converges almost everywhere to `C = lim (1/n) ∫ f_n`. -/ 34 axiom balancedKingman {Ω : Type*} [MeasurableSpace Ω] [StandardBorelSpace Ω] 35 (μ : Measure Ω) [IsProbabilityMeasure μ] [NullSingletonClass μ] 36 (σ : Ω ≃ᵐ Ω) (hσ : Ergodic σ μ) (f : ℕ → Ω → ℝ) (hf : IsSubadditiveFamily σ f) 37 (h1 : Integrable (fun ω => max (f 1 ω) 0) μ) : 38 ∃ C : EReal, C < ⊤ ∧ 39 (∀ᵐ ω ∂μ, Tendsto (fun n : ℕ => (f (2 * n) ((σ.symm : Ω → Ω)^[n] ω) / (2 * n) : EReal)) 40 atTop (𝓝 C)) ∧ 41 Tendsto (fun n : ℕ => Lax606786.LowerIntegrals.lowerIntegral μ (fun ω => f n ω / n)) atTop 42 (𝓝 C) 43 44 /-- The forward, shifted and balanced averages all converge to the same constant. -/ 45 axiom three_limits {Ω : Type*} [MeasurableSpace Ω] [StandardBorelSpace Ω] {μ : Measure Ω} 46 [IsProbabilityMeasure μ] [NullSingletonClass μ] {σ : Ω ≃ᵐ Ω} (hσ : Ergodic (⇑σ) μ) 47 {f : ℕ → Ω → ℝ} (hf : IsSubadditiveFamily (⇑σ) f) 48 (h1 : Integrable (fun ω => max (f 1 ω) 0) μ) (C : EReal) 49 (hC : ∀ᵐ ω ∂μ, Tendsto (fun n : ℕ => (f n ω / n : EReal)) atTop (𝓝 C)) : 50 (∀ᵐ ω ∂μ, Tendsto (fun n : ℕ => (f n (σ^[n] ω) / n : EReal)) atTop (𝓝 C)) ∧ 51 (∀ᵐ ω ∂μ, Tendsto (fun n : ℕ => 52 (f (2 * n) ((σ.symm : Ω → Ω)^[n] ω) / (2 * n) : EReal)) atTop (𝓝 C)) 53 54 end Lax606786.BalancedKingman 55 -
Linear combinations of weakly subadditive families
Let be an invertible ergodic measure-preserving transformation of a nonatomic standard probability space . Let be measurable and weakly subadditive: for constants and all ,
with integrable. Suppose and almost everywhere, with , and let , with if . Then satisfies, almost everywhere,
where is , or according as , or . This is Lemma 25 of Lee (2024).
1 import Mathlib.Dynamics.Ergodic.Ergodic 2 import Mathlib.MeasureTheory.Function.L1Space.Integrable 3 import Mathlib.MeasureTheory.Measure.Typeclasses.NullSingletonClass 4 import Mathlib.MeasureTheory.Constructions.Polish.Basic 5 import Mathlib.Data.EReal.Operations 6 … module docstring, 18 lines 25 26 namespace Lax606786.WeaklySubadditiveCombinations 27 28 open MeasureTheory Filter Topology 29 30 /-- The forward, shifted and balanced averages of `f + t g` converge to `C_f + t C_g`. -/ 31 axiom tendsto_add_smul {Ω : Type*} [MeasurableSpace Ω] [StandardBorelSpace Ω] 32 (μ : Measure Ω) [IsProbabilityMeasure μ] [NullSingletonClass μ] 33 (σ : Ω ≃ᵐ Ω) (hσ : Ergodic σ μ) 34 (f g : ℕ → Ω → ℝ) (Kf Kg : ℝ) 35 (hf : (∀ n, Measurable (f n)) ∧ ∀ m n ω, f (m + n) ω ≤ f m (σ^[n] ω) + f n ω + Kf) 36 (hg : (∀ n, Measurable (g n)) ∧ ∀ m n ω, g (m + n) ω ≤ g m (σ^[n] ω) + g n ω + Kg) 37 (hf1 : Integrable (fun ω => max (f 1 ω) 0) μ) 38 (hg1 : Integrable (fun ω => max (g 1 ω) 0) μ) 39 (t : ℝ) (Cf Cg : EReal) (hCf : Cf ≠ ⊤) (hCg : Cg ≠ ⊤) 40 (hdef : Cf = ⊥ → Cg = ⊥ → 0 ≤ t) 41 (hCfl : ∀ᵐ ω ∂μ, Tendsto (fun n : ℕ => (f n ω / n : EReal)) atTop (𝓝 Cf)) 42 (hCgl : ∀ᵐ ω ∂μ, Tendsto (fun n : ℕ => (g n ω / n : EReal)) atTop (𝓝 Cg)) : 43 ∀ᵐ ω ∂μ, 44 Tendsto (fun n : ℕ => ((f n ω + t * g n ω) / n : EReal)) atTop 45 (𝓝 (Cf + (t : EReal) * Cg)) ∧ 46 Tendsto (fun n : ℕ => ((f n (σ^[n] ω) + t * g n (σ^[n] ω)) / n : EReal)) atTop 47 (𝓝 (Cf + (t : EReal) * Cg)) ∧ 48 Tendsto (fun n : ℕ => 49 ((f (2 * n) ((σ.symm : Ω → Ω)^[n] ω) + t * g (2 * n) ((σ.symm : Ω → Ω)^[n] ω)) / 50 (2 * n) : EReal)) 51 atTop (𝓝 (Cf + (t : EReal) * Cg)) 52 53 end Lax606786.WeaklySubadditiveCombinations 54 -
Measurability of iterates, and constancy of the Lyapunov exponents
Let be a cocycle on a separable Banach space with its Borel -algebra.
- Each iterate is strongly measurable: is measurable for every and .
- For every the exponent is almost everywhere equal to a constant; so are the distinct exponents and their multiplicities .
- thm✓
Lax606786.CocycleBasics(1st statement) - thm✓
Lax606786.CocycleBasics(2nd statement) - thm✓
Lax606786.CocycleBasics(3rd statement) - thm✓
Lax606786.CocycleBasics(4th statement)
1 import Lax606786.Cocycles 2 … module docstring, 12 lines 15 16 namespace Lax606786.CocycleBasics 17 18 open MeasureTheory TopologicalSpace 19 open Lax606786.Cocycles 20 21 variable {Ω X : Type*} [MeasurableSpace Ω] [StandardBorelSpace Ω] 22 [NormedAddCommGroup X] [NormedSpace ℝ X] [MeasurableSpace X] 23 24 /-- The iterates are strongly measurable. -/ 25 axiom measurable_iterate_apply (R : Cocycle Ω X) [SeparableSpace X] [BorelSpace X] 26 (n : ℕ) (x : X) : Measurable fun ω => R.iterate n ω x 27 28 /-- `χ_k` is a.e. constant. -/ 29 axiom exists_chi_ae_eq (R : Cocycle Ω X) [SeparableSpace X] [BorelSpace X] (k : ℕ) : 30 ∃ c : EReal, ∀ᵐ ω ∂(R.μ), R.chi k ω = c 31 32 /-- The distinct exponent `λ_i` is a.e. constant. -/ 33 axiom exists_lambda_ae_eq (R : Cocycle Ω X) [SeparableSpace X] [BorelSpace X] (i : ℕ) : 34 ∃ c : EReal, ∀ᵐ ω ∂(R.μ), R.lambda i ω = c 35 36 /-- The multiplicity `m_i` is a.e. constant. -/ 37 axiom exists_mult_ae_eq (R : Cocycle Ω X) [SeparableSpace X] [BorelSpace X] (i : ℕ) : 38 ∃ m : ℕ, ∀ᵐ ω ∂(R.μ), R.mult i ω = m 39 40 end Lax606786.CocycleBasics 41 -
Existence of the top fast space
Let be a quasicompact cocycle on a separable Banach space with its Borel -algebra, whose top exponent is almost surely , with multiplicity .
Then there is a fast space of multiplicity at rate , on which the growth of is eventually positive.
This is Lemma 28 of Lee (2024).
1 import Lax606786.FastSpaces 2 … module docstring, 14 lines 17 18 namespace Lax606786.TopFastSpace 19 20 open MeasureTheory Filter TopologicalSpace 21 open Lax606786.Cocycles Lax606786.BanachGrassmannian Lax606786.OperatorStatistics Lax606786.FastSpaces 22 23 variable {Ω X : Type*} [MeasurableSpace Ω] [StandardBorelSpace Ω] 24 [NormedAddCommGroup X] [NormedSpace ℝ X] [MeasurableSpace X] 25 26 /-- The top fast space exists, with eventually positive growth. -/ 27 axiom exists_topFastSpace (R : Cocycle Ω X) [SeparableSpace X] [BorelSpace X] 28 [CompleteSpace X] (hqc : R.IsQuasicompact) 29 (lam : ℝ) (hlam : ∀ᵐ ω ∂(R.μ), R.chi 1 ω = (lam : EReal)) : 30 ∃ (m1 : ℕ) (_ : NeZero m1) (E : Ω → GrassmannianFin X m1), 31 (∀ᵐ ω ∂(R.μ), R.mult 1 ω = m1) ∧ IsFastSpace R m1 lam E ∧ 32 ∀ᵐ ω ∂(R.μ), ∀ᶠ n : ℕ in atTop, 0 < growth (R.iterate n ω) ((E ω).1 : Submodule ℝ X) 33 34 end Lax606786.TopFastSpace 35 -
The top-level fast/slow decomposition
Let be a quasicompact cocycle on a separable Banach space with its Borel -algebra, with top exponent of multiplicity and second exponent (possibly ), almost surely. Then there are
- a fast space of multiplicity at rate , with eventually positive growth;
- a measurable family of closed complements of of codimension , with ;
- strongly measurable, tempered idempotents with range and kernel containing ,
such that almost surely.
This is Lemma 29 of Lee (2024).
1 import Lax606786.FastSpaces 2 … module docstring, 19 lines 22 23 namespace Lax606786.TopDecomposition 24 25 open MeasureTheory Filter TopologicalSpace 26 open Lax606786.Cocycles Lax606786.BanachGrassmannian Lax606786.OperatorStatistics Lax606786.FastSpaces 27 Lax606786.ExtendedLog 28 29 /-- The top fast space, its slow complement and the slow projection. -/ 30 axiom topdecomp {Ω X : Type*} [MeasurableSpace Ω] [StandardBorelSpace Ω] 31 [NormedAddCommGroup X] [NormedSpace ℝ X] [MeasurableSpace X] 32 (R : Cocycle Ω X) [SeparableSpace X] [BorelSpace X] [CompleteSpace X] 33 (hqc : R.IsQuasicompact) 34 (lam : ℝ) (lam2 : EReal) (hlam : ∀ᵐ ω ∂(R.μ), R.chi 1 ω = (lam : EReal)) 35 (hlam2 : ∀ᵐ ω ∂(R.μ), R.lambda 2 ω = lam2) : 36 ∃ (m1 : ℕ) (_ : NeZero m1) (E : Ω → GrassmannianFin X m1) 37 (V : Ω → GrassmannianCofin X m1) (P : Ω → X →L[ℝ] X), 38 (∀ᵐ ω ∂(R.μ), R.mult 1 ω = m1) ∧ 39 IsFastSpace R m1 lam E ∧ 40 (∀ᵐ ω ∂(R.μ), ∀ᶠ n : ℕ in atTop, 41 0 < growth (R.iterate n ω) ((E ω).1 : Submodule ℝ X)) ∧ 42 (∀ x : X, Measurable fun ω => P ω x) ∧ 43 (∀ᵐ ω ∂(R.μ), (P ω).comp (P ω) = P ω) ∧ 44 (∀ᵐ ω ∂(R.μ), IsCompl ((E ω).1 : Submodule ℝ X) ((V ω).1.1 : Submodule ℝ X)) ∧ 45 (∀ᵐ ω ∂(R.μ), LinearMap.range (P ω : X →ₗ[ℝ] X) = ((V ω).1.1 : Submodule ℝ X) ∧ 46 ∀ e ∈ ((E ω).1 : Submodule ℝ X), P ω e = 0) ∧ 47 (∀ᵐ ω ∂(R.μ), Submodule.map (R.L ω : X →ₗ[ℝ] X) ((V ω).1.1 : Submodule ℝ X) 48 ≤ ((V (R.σ ω)).1.1 : Submodule ℝ X)) ∧ 49 R.IsTempered P ∧ 50 (∀ᵐ ω ∂(R.μ), Tendsto (fun n : ℕ => 51 logEReal ‖(R.iterate n ω).comp (P ω)‖ / (n : EReal)) atTop (nhds lam2)) 52 53 end Lax606786.TopDecomposition 54 -
Characterisation of the slow space
Let be a cocycle, a fast space of multiplicity at rate , and . Suppose are idempotents with range and kernel containing , where , and that almost surely. Then almost surely is the set of vectors growing at rate at most :
Compare Corollary 30 of Lee (2024).
1 import Lax606786.FastSpaces 2 … module docstring, 13 lines 16 17 namespace Lax606786.SlowSpaceCharacterisation 18 19 open MeasureTheory Filter 20 open Lax606786.Cocycles Lax606786.BanachGrassmannian Lax606786.FastSpaces Lax606786.ExtendedLog 21 22 /-- The slow space is the set of vectors with growth rate at most the slow rate. -/ 23 axiom V_characterisation {Ω X : Type*} [MeasurableSpace Ω] [StandardBorelSpace Ω] 24 [NormedAddCommGroup X] [NormedSpace ℝ X] [MeasurableSpace X] 25 (R : Cocycle Ω X) {m : ℕ} [NeZero m] {lam : ℝ} 26 {lam2 : EReal} (hgap : lam2 < lam) (E : Ω → GrassmannianFin X m) (hE : IsFastSpace R m lam E) 27 (V : Ω → GrassmannianCofin X m) (P : Ω → X →L[ℝ] X) 28 (hidem : ∀ᵐ ω ∂(R.μ), (P ω).comp (P ω) = P ω) 29 (hcompl : ∀ᵐ ω ∂(R.μ), IsCompl ((E ω).1 : Submodule ℝ X) ((V ω).1.1 : Submodule ℝ X)) 30 (hPV : ∀ᵐ ω ∂(R.μ), LinearMap.range (P ω : X →ₗ[ℝ] X) = ((V ω).1.1 : Submodule ℝ X) ∧ 31 ∀ e ∈ ((E ω).1 : Submodule ℝ X), P ω e = 0) 32 (hslow : ∀ᵐ ω ∂(R.μ), Tendsto (fun n : ℕ => 33 logEReal ‖(R.iterate n ω).comp (P ω)‖ / (n : EReal)) atTop (nhds lam2)) : 34 ∀ᵐ ω ∂(R.μ), (((V ω).1.1 : Submodule ℝ X) : Set X) = {x : X | R.lambdaAt ω x ≤ lam2} 35 36 end Lax606786.SlowSpaceCharacterisation 37 -
The construction in the proof of Theorem 3 of Lee (2024): the decomposition is assembled level by level from the base level and the inductive step, each fast space being the top fast space of the cocycle restricted to the previous slow space.
-
Gelfand numbers, Bernstein numbers and the compactness seminorm
Let be a real normed space, a bounded operator on and . Then
The first inequality holds for every . These are Lemmas 31–33 of Lee (2024). Sharper constants follow from the Kadec–Snobar theorem (Kadec and Snobar 1971; see also Wojtaszczyk 1991, III.B.10) and from Pietsch's theory of -numbers (Pietsch 1974); any constant depending on alone suffices for comparing exponential growth rates.
Gelfand numbers are nonnegative.
- thm✓
Lax606786.GelfandBernsteinComparison(1st statement) - thm✓
Lax606786.GelfandBernsteinComparison(2nd statement) - thm✓
Lax606786.GelfandBernsteinComparison(3rd statement) - thm✓
Lax606786.GelfandBernsteinComparison(4th statement)
1 import Lax606786.OperatorStatistics 2 … module docstring, 15 lines 18 19 namespace Lax606786.GelfandBernsteinComparison 20 21 open Lax606786.OperatorStatistics 22 23 variable {X : Type*} [NormedAddCommGroup X] [NormedSpace ℝ X] 24 25 /-- `ρ_k ≤ s_k`. -/ 26 axiom bernsteinNumber_le_gelfandNumber (T : X →L[ℝ] X) (k : ℕ) : 27 bernsteinNumber T k ≤ gelfandNumber T k 28 29 /-- `s_k ≤ 4^{k-1} (k-1)! ρ_k`. -/ 30 axiom gelfandNumber_le_bernsteinNumber (T : X →L[ℝ] X) {k : ℕ} (hk : 1 ≤ k) : 31 gelfandNumber T k ≤ 4 ^ (k - 1) * (Nat.factorial (k - 1) : ℝ) * bernsteinNumber T k 32 33 /-- `‖T‖_c ≤ 2k s_k`. -/ 34 axiom compactSeminorm_le_gelfandNumber (T : X →L[ℝ] X) {k : ℕ} (hk : 1 ≤ k) : 35 compactSeminorm T ≤ 2 * (k : ℝ) * gelfandNumber T k 36 37 /-- Gelfand numbers are nonnegative. -/ 38 axiom gelfandNumber_nonneg (T : X →L[ℝ] X) (k : ℕ) : 0 ≤ gelfandNumber T k 39 40 end Lax606786.GelfandBernsteinComparison 41 - thm✓
-
The Bernstein index is the index of compactness
Let be a cocycle on a separable Banach space . The growth rate of the compactness seminorm (the distance to the compact operators) of the iterates exists almost everywhere and equals the Bernstein index :
Since is almost everywhere constant, the index of compactness always exists as an almost-everywhere constant, and almost everywhere. This is Lemma 34 of Lee (2024), which assumes that the limit exists; here its existence is part of the statement.
- thm✓
Lax606786.IndexOfCompactness(1st statement) - thm✓
Lax606786.IndexOfCompactness(2nd statement) - thm✓
Lax606786.IndexOfCompactness(3rd statement)
1 import Lax606786.Cocycles 2 … module docstring, 14 lines 17 18 namespace Lax606786.IndexOfCompactness 19 20 open MeasureTheory Filter TopologicalSpace 21 open Lax606786.OperatorStatistics Lax606786.ExtendedLog Lax606786.Cocycles 22 23 /-- `ν = κ` almost everywhere, where `κ` is the growth rate of `‖𝓛^{(n)}_ω‖_c`. -/ 24 axiom nu_eq_compactnessIndex {Ω X : Type*} [MeasurableSpace Ω] [StandardBorelSpace Ω] 25 [NormedAddCommGroup X] [NormedSpace ℝ X] [MeasurableSpace X] 26 (R : Cocycle Ω X) [SeparableSpace X] [BorelSpace X] 27 (κ : EReal) (hκ : ∀ᵐ ω ∂(R.μ), Tendsto (fun n : ℕ => 28 logEReal (compactSeminorm (R.iterate n ω)) / (n : EReal)) atTop (nhds κ)) : 29 ∀ᵐ ω ∂(R.μ), R.nu ω = κ 30 31 /-- `(1/n) log ‖𝓛^{(n)}_ω‖_c → ν(ω)` almost everywhere. -/ 32 axiom ae_tendsto_compactSeminorm {Ω X : Type*} [MeasurableSpace Ω] [StandardBorelSpace Ω] 33 [NormedAddCommGroup X] [NormedSpace ℝ X] [MeasurableSpace X] 34 (R : Cocycle Ω X) [SeparableSpace X] [BorelSpace X] : 35 ∀ᵐ ω ∂(R.μ), Tendsto (fun n : ℕ => 36 logEReal (compactSeminorm (R.iterate n ω)) / (n : EReal)) atTop (nhds (R.nu ω)) 37 38 /-- The index of compactness exists: `(1/n) log ‖𝓛^{(n)}_ω‖_c` converges almost everywhere to a 39 constant. -/ 40 axiom exists_compactnessIndex {Ω X : Type*} [MeasurableSpace Ω] [StandardBorelSpace Ω] 41 [NormedAddCommGroup X] [NormedSpace ℝ X] [MeasurableSpace X] 42 (R : Cocycle Ω X) [SeparableSpace X] [BorelSpace X] : 43 ∃ κ : EReal, ∀ᵐ ω ∂(R.μ), Tendsto (fun n : ℕ => 44 logEReal (compactSeminorm (R.iterate n ω)) / (n : EReal)) atTop (nhds κ) 45 46 end Lax606786.IndexOfCompactness 47 - thm✓
-
Slow spaces on a separable Banach space need not be measurable
Let be the circle with Lebesgue measure, rotated by the irrational angle : . Let with its Borel -algebra, and its -th unit vector. For and let if the representative of in lies in , and otherwise, and put
Then is separable, and is a quasicompact cocycle over with and . For every its slow space
belongs to the Grassmannian , yet there is no measurable map with equal to this slow space for every . So separability of alone does not make the slow spaces of the Oseledets decomposition measurable.
The mechanism is that of Horan (2020, Remark 4.5.7), who observes that for a family of Perron–Frobenius operators on at mutual distance at least , measurability of would make arbitrary sets of base points measurable. Lee (2024, Example 35) gives a version over a full shift, on with summing from ; that functional vanishes on , so here the sum runs over all coordinates.
1 import Lax606786.BanachGrassmannian 2 import Lax606786.Cocycles 3 import Mathlib.Analysis.Normed.Lp.lpSpace 4 import Mathlib.MeasureTheory.Constructions.Polish.Basic 5 import Mathlib.MeasureTheory.Integral.IntervalIntegral.Periodic 6 … module docstring, 27 lines 34 35 namespace Lax606786.SlowSpaceNonmeasurability 36 37 open MeasureTheory 38 open Lax606786.BanachGrassmannian Lax606786.Cocycles 39 40 /-- `ℓ¹(ℤ)` with its Borel `σ`-algebra. -/ 41 noncomputable instance instMeasurableSpaceL1 : MeasurableSpace (lp (fun _ : ℤ => ℝ) 1) := 42 borel _ 43 44 instance instBorelSpaceL1 : BorelSpace (lp (fun _ : ℤ => ℝ) 1) := ⟨rfl⟩ 45 46 /-- The rotation angle `α = √2 mod 1`. -/ 47 noncomputable def angle : AddCircle (1 : ℝ) := ((Real.sqrt 2 : ℝ) : AddCircle (1 : ℝ)) 48 49 /-- `s_i(ω) = 1` if the representative of `ω + iα` in `[0, 1)` lies in `[0, 1/2)`, else `-1`. -/ 50 noncomputable def sign (ω : AddCircle (1 : ℝ)) (i : ℤ) : ℝ := 51 if ((AddCircle.equivIco 1 0 (ω + i • angle) : ℝ)) < 1 / 2 then 1 else -1 52 53 /-- `ℓ¹(ℤ)` is separable, and the cocycle `𝓛_ω x = (Σᵢ sᵢ(ω) xᵢ) e₀` over the rotation by `α` 54 of the circle with Lebesgue measure is quasicompact; its slow space `{x | λ_ω(x) ≤ λ₂(ω)}` is 55 `ker φ_ω` and lies in `𝒢X` for every `ω`, but no measurable map `Ω → 𝒢X` takes it as value at 56 every `ω`. -/ 57 axiom exists_cocycle_slowSpace_not_measurable : 58 TopologicalSpace.SeparableSpace (lp (fun _ : ℤ => ℝ) 1) ∧ 59 ∃ R : Cocycle (AddCircle (1 : ℝ)) (lp (fun _ : ℤ => ℝ) 1), 60 R.μ = volume ∧ (∀ ω, R.σ ω = ω + angle) ∧ 61 (∀ ω x, R.L ω x = (∑' i, sign ω i * x i) • lp.single 1 0 (1 : ℝ)) ∧ 62 R.IsQuasicompact ∧ 63 (∀ ω, {x | R.lambdaAt ω x ≤ R.lambda 2 ω} = {x | ∑' i, sign ω i * x i = 0}) ∧ 64 (∀ ω, ∃ W : Grassmannian (lp (fun _ : ℤ => ℝ) 1), 65 ((W.1 : Submodule ℝ _) : Set _) = {x | R.lambdaAt ω x ≤ R.lambda 2 ω}) ∧ 66 ¬ ∃ W : AddCircle (1 : ℝ) → Grassmannian (lp (fun _ : ℤ => ℝ) 1), Measurable W ∧ 67 ∀ ω, (((W ω).1 : Submodule ℝ _) : Set _) = {x | R.lambdaAt ω x ≤ R.lambda 2 ω} 68 69 end Lax606786.SlowSpaceNonmeasurability 70
Loading the paper…