-
Notifications
You must be signed in to change notification settings - Fork 276
/
Extension.lean
192 lines (161 loc) · 8.64 KB
/
Extension.lean
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
/-
Copyright (c) 2020 Yury Kudryashov All rights reserved.
Released under Apache 2.0 license as described in the file LICENSE.
Authors: Yury Kudryashov, Heather Macbeth
-/
import Mathlib.Analysis.Convex.Cone.Extension
import Mathlib.Analysis.NormedSpace.RCLike
import Mathlib.Analysis.NormedSpace.Extend
import Mathlib.Analysis.RCLike.Lemmas
#align_import analysis.normed_space.hahn_banach.extension from "leanprover-community/mathlib"@"915591b2bb3ea303648db07284a161a7f2a9e3d4"
/-!
# Extension Hahn-Banach theorem
In this file we prove the analytic Hahn-Banach theorem. For any continuous linear function on a
subspace, we can extend it to a function on the entire space without changing its norm.
We prove
* `Real.exists_extension_norm_eq`: Hahn-Banach theorem for continuous linear functions on normed
spaces over `ℝ`.
* `exists_extension_norm_eq`: Hahn-Banach theorem for continuous linear functions on normed spaces
over `ℝ` or `ℂ`.
In order to state and prove the corollaries uniformly, we prove the statements for a field `𝕜`
satisfying `RCLike 𝕜`.
In this setting, `exists_dual_vector` states that, for any nonzero `x`, there exists a continuous
linear form `g` of norm `1` with `g x = ‖x‖` (where the norm has to be interpreted as an element
of `𝕜`).
-/
universe u v
namespace Real
variable {E : Type*} [SeminormedAddCommGroup E] [NormedSpace ℝ E]
/-- **Hahn-Banach theorem** for continuous linear functions over `ℝ`.
See also `exists_extension_norm_eq` in the root namespace for a more general version
that works both for `ℝ` and `ℂ`. -/
theorem exists_extension_norm_eq (p : Subspace ℝ E) (f : p →L[ℝ] ℝ) :
∃ g : E →L[ℝ] ℝ, (∀ x : p, g x = f x) ∧ ‖g‖ = ‖f‖ := by
rcases exists_extension_of_le_sublinear ⟨p, f⟩ (fun x => ‖f‖ * ‖x‖)
(fun c hc x => by simp only [norm_smul c x, Real.norm_eq_abs, abs_of_pos hc, mul_left_comm])
(fun x y => by -- Porting note: placeholder filled here
rw [← left_distrib]
exact mul_le_mul_of_nonneg_left (norm_add_le x y) (@norm_nonneg _ _ f))
fun x => le_trans (le_abs_self _) (f.le_opNorm _) with ⟨g, g_eq, g_le⟩
set g' :=
g.mkContinuous ‖f‖ fun x => abs_le.2 ⟨neg_le.1 <| g.map_neg x ▸ norm_neg x ▸ g_le (-x), g_le x⟩
refine' ⟨g', g_eq, _⟩
apply le_antisymm (g.mkContinuous_norm_le (norm_nonneg f) _)
refine' f.opNorm_le_bound (norm_nonneg _) fun x => _
dsimp at g_eq
rw [← g_eq]
apply g'.le_opNorm
#align real.exists_extension_norm_eq Real.exists_extension_norm_eq
end Real
section RCLike
open RCLike
variable {𝕜 : Type*} [RCLike 𝕜] {E F : Type*}
[SeminormedAddCommGroup E] [NormedSpace 𝕜 E]
[NormedAddCommGroup F] [NormedSpace 𝕜 F]
/-- **Hahn-Banach theorem** for continuous linear functions over `𝕜` satisfying `RCLike 𝕜`. -/
theorem exists_extension_norm_eq (p : Subspace 𝕜 E) (f : p →L[𝕜] 𝕜) :
∃ g : E →L[𝕜] 𝕜, (∀ x : p, g x = f x) ∧ ‖g‖ = ‖f‖ := by
letI : Module ℝ E := RestrictScalars.module ℝ 𝕜 E
letI : IsScalarTower ℝ 𝕜 E := RestrictScalars.isScalarTower _ _ _
letI : NormedSpace ℝ E := NormedSpace.restrictScalars _ 𝕜 _
-- Let `fr: p →L[ℝ] ℝ` be the real part of `f`.
let fr := reCLM.comp (f.restrictScalars ℝ)
-- Use the real version to get a norm-preserving extension of `fr`, which
-- we'll call `g : E →L[ℝ] ℝ`.
rcases Real.exists_extension_norm_eq (p.restrictScalars ℝ) fr with ⟨g, ⟨hextends, hnormeq⟩⟩
-- Now `g` can be extended to the `E →L[𝕜] 𝕜` we need.
refine' ⟨g.extendTo𝕜, _⟩
-- It is an extension of `f`.
have h : ∀ x : p, g.extendTo𝕜 x = f x := by
intro x
-- This used to be `rw`, but we need `erw` after leanprover/lean4#2644
erw [ContinuousLinearMap.extendTo𝕜_apply, ← Submodule.coe_smul, hextends, hextends]
have : (fr x : 𝕜) - I * ↑(fr (I • x)) = (re (f x) : 𝕜) - (I : 𝕜) * re (f ((I : 𝕜) • x)) := by
rfl
-- This used to be `rw`, but we need `erw` after leanprover/lean4#2644
erw [this]
apply ext
· simp only [add_zero, Algebra.id.smul_eq_mul, I_re, ofReal_im, AddMonoidHom.map_add, zero_sub,
I_im', zero_mul, ofReal_re, eq_self_iff_true, sub_zero, mul_neg, ofReal_neg,
mul_re, mul_zero, sub_neg_eq_add, ContinuousLinearMap.map_smul]
· simp only [Algebra.id.smul_eq_mul, I_re, ofReal_im, AddMonoidHom.map_add, zero_sub, I_im',
zero_mul, ofReal_re, mul_neg, mul_im, zero_add, ofReal_neg, mul_re,
sub_neg_eq_add, ContinuousLinearMap.map_smul]
-- And we derive the equality of the norms by bounding on both sides.
refine' ⟨h, le_antisymm _ _⟩
· calc
‖g.extendTo𝕜‖ = ‖g‖ := g.norm_extendTo𝕜
_ = ‖fr‖ := hnormeq
_ ≤ ‖reCLM‖ * ‖f‖ := ContinuousLinearMap.opNorm_comp_le _ _
_ = ‖f‖ := by rw [reCLM_norm, one_mul]
· exact f.opNorm_le_bound g.extendTo𝕜.opNorm_nonneg fun x => h x ▸ g.extendTo𝕜.le_opNorm x
#align exists_extension_norm_eq exists_extension_norm_eq
open FiniteDimensional
/-- Corollary of the **Hahn-Banach theorem**: if `f : p → F` is a continuous linear map
from a submodule of a normed space `E` over `𝕜`, `𝕜 = ℝ` or `𝕜 = ℂ`,
with a finite dimensional range,
then `f` admits an extension to a continuous linear map `E → F`.
Note that contrary to the case `F = 𝕜`, see `exists_extension_norm_eq`,
we provide no estimates on the norm of the extension.
-/
lemma ContinuousLinearMap.exist_extension_of_finiteDimensional_range {p : Submodule 𝕜 E}
(f : p →L[𝕜] F) [FiniteDimensional 𝕜 (LinearMap.range f)] :
∃ g : E →L[𝕜] F, f = g.comp p.subtypeL := by
set b := finBasis 𝕜 (LinearMap.range f)
set e := b.equivFunL
set fi := fun i ↦ (LinearMap.toContinuousLinearMap (b.coord i)).comp
(f.codRestrict _ <| LinearMap.mem_range_self _)
choose gi hgf _ using fun i ↦ exists_extension_norm_eq p (fi i)
use (LinearMap.range f).subtypeL.comp <| e.symm.toContinuousLinearMap.comp (.pi gi)
ext x
simp [fi, e, hgf]
/-- A finite dimensional submodule over `ℝ` or `ℂ` is `Submodule.ClosedComplemented`. -/
lemma Submodule.ClosedComplemented.of_finiteDimensional (p : Submodule 𝕜 F)
[FiniteDimensional 𝕜 p] : p.ClosedComplemented :=
let ⟨g, hg⟩ := (ContinuousLinearMap.id 𝕜 p).exist_extension_of_finiteDimensional_range
⟨g, DFunLike.congr_fun hg.symm⟩
end RCLike
section DualVector
variable (𝕜 : Type v) [RCLike 𝕜]
variable {E : Type u} [NormedAddCommGroup E] [NormedSpace 𝕜 E]
open ContinuousLinearEquiv Submodule
open scoped Classical
theorem coord_norm' {x : E} (h : x ≠ 0) : ‖(‖x‖ : 𝕜) • coord 𝕜 x h‖ = 1 := by
rw [norm_smul (α := 𝕜) (x := coord 𝕜 x h), RCLike.norm_coe_norm, coord_norm,
mul_inv_cancel (mt norm_eq_zero.mp h)]
#align coord_norm' coord_norm'
/-- Corollary of Hahn-Banach. Given a nonzero element `x` of a normed space, there exists an
element of the dual space, of norm `1`, whose value on `x` is `‖x‖`. -/
theorem exists_dual_vector (x : E) (h : x ≠ 0) : ∃ g : E →L[𝕜] 𝕜, ‖g‖ = 1 ∧ g x = ‖x‖ := by
let p : Submodule 𝕜 E := 𝕜 ∙ x
let f := (‖x‖ : 𝕜) • coord 𝕜 x h
obtain ⟨g, hg⟩ := exists_extension_norm_eq p f
refine' ⟨g, _, _⟩
· rw [hg.2, coord_norm']
· calc
g x = g (⟨x, mem_span_singleton_self x⟩ : 𝕜 ∙ x) := by rw [coe_mk]
_ = ((‖x‖ : 𝕜) • coord 𝕜 x h) (⟨x, mem_span_singleton_self x⟩ : 𝕜 ∙ x) := by rw [← hg.1]
_ = ‖x‖ := by simp
#align exists_dual_vector exists_dual_vector
/-- Variant of Hahn-Banach, eliminating the hypothesis that `x` be nonzero, and choosing
the dual element arbitrarily when `x = 0`. -/
theorem exists_dual_vector' [Nontrivial E] (x : E) : ∃ g : E →L[𝕜] 𝕜, ‖g‖ = 1 ∧ g x = ‖x‖ := by
by_cases hx : x = 0
· obtain ⟨y, hy⟩ := exists_ne (0 : E)
obtain ⟨g, hg⟩ : ∃ g : E →L[𝕜] 𝕜, ‖g‖ = 1 ∧ g y = ‖y‖ := exists_dual_vector 𝕜 y hy
refine' ⟨g, hg.left, _⟩
simp [hx]
· exact exists_dual_vector 𝕜 x hx
#align exists_dual_vector' exists_dual_vector'
/-- Variant of Hahn-Banach, eliminating the hypothesis that `x` be nonzero, but only ensuring that
the dual element has norm at most `1` (this can not be improved for the trivial
vector space). -/
theorem exists_dual_vector'' (x : E) : ∃ g : E →L[𝕜] 𝕜, ‖g‖ ≤ 1 ∧ g x = ‖x‖ := by
by_cases hx : x = 0
· refine' ⟨0, by simp, _⟩
symm
simp [hx]
· rcases exists_dual_vector 𝕜 x hx with ⟨g, g_norm, g_eq⟩
exact ⟨g, g_norm.le, g_eq⟩
#align exists_dual_vector'' exists_dual_vector''
end DualVector