Datasets:

Modalities:
Text
Languages:
English
Libraries:
Datasets
License:
Zhangir Azerbayev
squashed?
4365a98
raw
history blame
21.6 kB
/-
Copyright (c) 2021 Jireh Loreaux. All rights reserved.
Released under Apache 2.0 license as described in the file LICENSE.
Authors: Jireh Loreaux
-/
import algebra.algebra.spectrum
import analysis.special_functions.pow
import analysis.special_functions.exponential
import analysis.complex.liouville
import analysis.analytic.radius_liminf
/-!
# The spectrum of elements in a complete normed algebra
This file contains the basic theory for the resolvent and spectrum of a Banach algebra.
## Main definitions
* `spectral_radius : ℝβ‰₯0∞`: supremum of `βˆ₯kβˆ₯β‚Š` for all `k ∈ spectrum π•œ a`
## Main statements
* `spectrum.is_open_resolvent_set`: the resolvent set is open.
* `spectrum.is_closed`: the spectrum is closed.
* `spectrum.subset_closed_ball_norm`: the spectrum is a subset of closed disk of radius
equal to the norm.
* `spectrum.is_compact`: the spectrum is compact.
* `spectrum.spectral_radius_le_nnnorm`: the spectral radius is bounded above by the norm.
* `spectrum.has_deriv_at_resolvent`: the resolvent function is differentiable on the resolvent set.
* `spectrum.pow_nnnorm_pow_one_div_tendsto_nhds_spectral_radius`: Gelfand's formula for the
spectral radius in Banach algebras over `β„‚`.
* `spectrum.nonempty`: the spectrum of any element in a complex Banach algebra is nonempty.
* `normed_division_ring.alg_equiv_complex_of_complete`: **Gelfand-Mazur theorem** For a complex
Banach division algebra, the natural `algebra_map β„‚ A` is an algebra isomorphism whose inverse
is given by selecting the (unique) element of `spectrum β„‚ a`
## TODO
* compute all derivatives of `resolvent a`.
-/
open_locale ennreal
/-- The *spectral radius* is the supremum of the `nnnorm` (`βˆ₯⬝βˆ₯β‚Š`) of elements in the spectrum,
coerced into an element of `ℝβ‰₯0∞`. Note that it is possible for `spectrum π•œ a = βˆ…`. In this
case, `spectral_radius a = 0`. It is also possible that `spectrum π•œ a` be unbounded (though
not for Banach algebras, see `spectrum.is_bounded`, below). In this case,
`spectral_radius a = ∞`. -/
noncomputable def spectral_radius (π•œ : Type*) {A : Type*} [normed_field π•œ] [ring A]
[algebra π•œ A] (a : A) : ℝβ‰₯0∞ :=
⨆ k ∈ spectrum π•œ a, βˆ₯kβˆ₯β‚Š
variables {π•œ : Type*} {A : Type*}
namespace spectrum
section spectrum_compact
variables [normed_field π•œ] [normed_ring A] [normed_algebra π•œ A]
local notation `Οƒ` := spectrum π•œ
local notation `ρ` := resolvent_set π•œ
local notation `↑ₐ` := algebra_map π•œ A
lemma mem_resolvent_set_of_spectral_radius_lt {a : A} {k : π•œ} (h : spectral_radius π•œ a < βˆ₯kβˆ₯β‚Š) :
k ∈ ρ a :=
not_not.mp $ Ξ» hn, h.not_le $ le_suprβ‚‚ k hn
variable [complete_space A]
lemma is_open_resolvent_set (a : A) : is_open (ρ a) :=
units.is_open.preimage ((algebra_map_clm π•œ A).continuous.sub continuous_const)
lemma is_closed (a : A) : is_closed (Οƒ a) :=
(is_open_resolvent_set a).is_closed_compl
lemma mem_resolvent_of_norm_lt [norm_one_class A] {a : A} {k : π•œ} (h : βˆ₯aβˆ₯ < βˆ₯kβˆ₯) :
k ∈ ρ a :=
begin
rw [resolvent_set, set.mem_set_of_eq, algebra.algebra_map_eq_smul_one],
have hk : k β‰  0 := ne_zero_of_norm_ne_zero (by linarith [norm_nonneg a]),
let ku := units.map (↑ₐ).to_monoid_hom (units.mk0 k hk),
have hku : βˆ₯-aβˆ₯ < βˆ₯(↑ku⁻¹:A)βˆ₯⁻¹ := by simpa [ku, algebra_map_isometry] using h,
simpa [ku, sub_eq_add_neg, algebra.algebra_map_eq_smul_one] using (ku.add (-a) hku).is_unit,
end
lemma norm_le_norm_of_mem [norm_one_class A] {a : A} {k : π•œ} (hk : k ∈ Οƒ a) :
βˆ₯kβˆ₯ ≀ βˆ₯aβˆ₯ :=
le_of_not_lt $ mt mem_resolvent_of_norm_lt hk
lemma subset_closed_ball_norm [norm_one_class A] (a : A) :
Οƒ a βŠ† metric.closed_ball (0 : π•œ) (βˆ₯aβˆ₯) :=
Ξ» k hk, by simp [norm_le_norm_of_mem hk]
lemma is_bounded [norm_one_class A] (a : A) : metric.bounded (Οƒ a) :=
(metric.bounded_iff_subset_ball 0).mpr ⟨βˆ₯aβˆ₯, subset_closed_ball_norm a⟩
theorem is_compact [norm_one_class A] [proper_space π•œ] (a : A) : is_compact (Οƒ a) :=
metric.is_compact_of_is_closed_bounded (is_closed a) (is_bounded a)
theorem spectral_radius_le_nnnorm [norm_one_class A] (a : A) :
spectral_radius π•œ a ≀ βˆ₯aβˆ₯β‚Š :=
by { refine suprβ‚‚_le (Ξ» k hk, _), exact_mod_cast norm_le_norm_of_mem hk }
open ennreal polynomial
variable (π•œ)
theorem spectral_radius_le_pow_nnnorm_pow_one_div [norm_one_class A] (a : A) (n : β„•) :
spectral_radius π•œ a ≀ βˆ₯a ^ (n + 1)βˆ₯β‚Š ^ (1 / (n + 1) : ℝ) :=
begin
refine suprβ‚‚_le (Ξ» k hk, _),
/- apply easy direction of the spectral mapping theorem for polynomials -/
have pow_mem : k ^ (n + 1) ∈ Οƒ (a ^ (n + 1)),
by simpa only [one_mul, algebra.algebra_map_eq_smul_one, one_smul, aeval_monomial, one_mul,
eval_monomial] using subset_polynomial_aeval a (monomial (n + 1) (1 : π•œ)) ⟨k, hk, rfl⟩,
/- power of the norm is bounded by norm of the power -/
have nnnorm_pow_le : (↑(βˆ₯kβˆ₯β‚Š ^ (n + 1)) : ℝβ‰₯0∞) ≀ ↑βˆ₯a ^ (n + 1)βˆ₯β‚Š,
by simpa only [norm_to_nnreal, nnnorm_pow k (n+1)]
using coe_mono (real.to_nnreal_mono (norm_le_norm_of_mem pow_mem)),
/- take (n + 1)α΅—Κ° roots and clean up the left-hand side -/
have hn : 0 < ((n + 1 : β„•) : ℝ), by exact_mod_cast nat.succ_pos',
convert monotone_rpow_of_nonneg (one_div_pos.mpr hn).le nnnorm_pow_le,
erw [coe_pow, ←rpow_nat_cast, ←rpow_mul, mul_one_div_cancel hn.ne', rpow_one], rw nat.cast_succ,
end
end spectrum_compact
section resolvent
open filter asymptotics
variables [nontrivially_normed_field π•œ] [normed_ring A] [normed_algebra π•œ A] [complete_space A]
local notation `ρ` := resolvent_set π•œ
local notation `↑ₐ` := algebra_map π•œ A
theorem has_deriv_at_resolvent {a : A} {k : π•œ} (hk : k ∈ ρ a) :
has_deriv_at (resolvent a) (-(resolvent a k) ^ 2) k :=
begin
have H₁ : has_fderiv_at ring.inverse _ (↑ₐk - a) := has_fderiv_at_ring_inverse hk.unit,
have Hβ‚‚ : has_deriv_at (Ξ» k, ↑ₐk - a) 1 k,
{ simpa using (algebra.linear_map π•œ A).has_deriv_at.sub_const a },
simpa [resolvent, sq, hk.unit_spec, ← ring.inverse_unit hk.unit] using H₁.comp_has_deriv_at k Hβ‚‚,
end
/- TODO: Once there is sufficient API for bornology, we should get a nice filter / asymptotics
version of this, for example: `tendsto (resolvent a) (cobounded π•œ) (𝓝 0)` or more specifically
`(resolvent a) =O[cobounded π•œ] (Ξ» z, z⁻¹)`. -/
lemma norm_resolvent_le_forall (a : A) :
βˆ€ Ξ΅ > 0, βˆƒ R > 0, βˆ€ z : π•œ, R ≀ βˆ₯zβˆ₯ β†’ βˆ₯resolvent a zβˆ₯ ≀ Ξ΅ :=
begin
obtain ⟨c, c_pos, hc⟩ := (@normed_ring.inverse_one_sub_norm A _ _).exists_pos,
rw [is_O_with_iff, eventually_iff, metric.mem_nhds_iff] at hc,
rcases hc with ⟨δ, δ_pos, hδ⟩,
simp only [cstar_ring.norm_one, mul_one] at hΞ΄,
intros Ξ΅ hΞ΅,
have ha₁ : 0 < βˆ₯aβˆ₯ + 1 := lt_of_le_of_lt (norm_nonneg a) (lt_add_one _),
have min_pos : 0 < min (Ξ΄ * (βˆ₯aβˆ₯ + 1)⁻¹) (Ξ΅ * c⁻¹),
from lt_min (mul_pos Ξ΄_pos (inv_pos.mpr ha₁)) (mul_pos hΞ΅ (inv_pos.mpr c_pos)),
refine ⟨(min (Ξ΄ * (βˆ₯aβˆ₯ + 1)⁻¹) (Ξ΅ * c⁻¹))⁻¹, inv_pos.mpr min_pos, (Ξ» z hz, _)⟩,
have hnz : z β‰  0 := norm_pos_iff.mp (lt_of_lt_of_le (inv_pos.mpr min_pos) hz),
replace hz := inv_le_of_inv_le min_pos hz,
rcases (⟨units.mk0 z hnz, units.coe_mk0 hnz⟩ : is_unit z) with ⟨z, rfl⟩,
have lt_Ξ΄ : βˆ₯z⁻¹ β€’ aβˆ₯ < Ξ΄,
{ rw [units.smul_def, norm_smul, units.coe_inv, norm_inv],
calc βˆ₯(z : π•œ)βˆ₯⁻¹ * βˆ₯aβˆ₯ ≀ Ξ΄ * (βˆ₯aβˆ₯ + 1)⁻¹ * βˆ₯aβˆ₯
: mul_le_mul_of_nonneg_right (hz.trans (min_le_left _ _)) (norm_nonneg _)
... < Ξ΄
: by { conv { rw mul_assoc, to_rhs, rw (mul_one Ξ΄).symm },
exact mul_lt_mul_of_pos_left
((inv_mul_lt_iff ha₁).mpr ((mul_one (βˆ₯aβˆ₯ + 1)).symm β–Έ (lt_add_one _))) Ξ΄_pos } },
rw [←inv_smul_smul z (resolvent a (z : π•œ)), units_smul_resolvent_self, resolvent,
algebra.algebra_map_eq_smul_one, one_smul, units.smul_def, norm_smul, units.coe_inv, norm_inv],
calc _ ≀ Ξ΅ * c⁻¹ * c : mul_le_mul (hz.trans (min_le_right _ _)) (hΞ΄ (mem_ball_zero_iff.mpr lt_Ξ΄))
(norm_nonneg _) (mul_pos hΞ΅ (inv_pos.mpr c_pos)).le
... = _ : inv_mul_cancel_rightβ‚€ c_pos.ne.symm Ξ΅,
end
end resolvent
section one_sub_smul
open continuous_multilinear_map ennreal formal_multilinear_series
open_locale nnreal ennreal
variables
[nontrivially_normed_field π•œ] [normed_ring A] [normed_algebra π•œ A]
variable (π•œ)
/-- In a Banach algebra `A` over a nontrivially normed field `π•œ`, for any `a : A` the
power series with coefficients `a ^ n` represents the function `(1 - z β€’ a)⁻¹` in a disk of
radius `βˆ₯aβˆ₯β‚Šβ»ΒΉ`. -/
lemma has_fpower_series_on_ball_inverse_one_sub_smul [complete_space A] (a : A) :
has_fpower_series_on_ball (Ξ» z : π•œ, ring.inverse (1 - z β€’ a))
(Ξ» n, continuous_multilinear_map.mk_pi_field π•œ (fin n) (a ^ n)) 0 (βˆ₯aβˆ₯β‚Š)⁻¹ :=
{ r_le :=
begin
refine le_of_forall_nnreal_lt (Ξ» r hr, le_radius_of_bound_nnreal _ (max 1 βˆ₯(1 : A)βˆ₯β‚Š) (Ξ» n, _)),
rw [←norm_to_nnreal, norm_mk_pi_field, norm_to_nnreal],
cases n,
{ simp only [le_refl, mul_one, or_true, le_max_iff, pow_zero] },
{ refine le_trans (le_trans (mul_le_mul_right' (nnnorm_pow_le' a n.succ_pos) (r ^ n.succ)) _)
(le_max_left _ _),
{ by_cases βˆ₯aβˆ₯β‚Š = 0,
{ simp only [h, zero_mul, zero_le', pow_succ], },
{ rw [←coe_inv h, coe_lt_coe, nnreal.lt_inv_iff_mul_lt h] at hr,
simpa only [←mul_pow, mul_comm] using pow_le_one' hr.le n.succ } } }
end,
r_pos := ennreal.inv_pos.mpr coe_ne_top,
has_sum := Ξ» y hy,
begin
have norm_lt : βˆ₯y β€’ aβˆ₯ < 1,
{ by_cases h : βˆ₯aβˆ₯β‚Š = 0,
{ simp only [nnnorm_eq_zero.mp h, norm_zero, zero_lt_one, smul_zero] },
{ have nnnorm_lt : βˆ₯yβˆ₯β‚Š < βˆ₯aβˆ₯β‚Šβ»ΒΉ,
by simpa only [←coe_inv h, mem_ball_zero_iff, metric.emetric_ball_nnreal] using hy,
rwa [←coe_nnnorm, ←real.lt_to_nnreal_iff_coe_lt, real.to_nnreal_one, nnnorm_smul,
←nnreal.lt_inv_iff_mul_lt h] } },
simpa [←smul_pow, (normed_ring.summable_geometric_of_norm_lt_1 _ norm_lt).has_sum_iff]
using (normed_ring.inverse_one_sub _ norm_lt).symm,
end }
variable {π•œ}
lemma is_unit_one_sub_smul_of_lt_inv_radius {a : A} {z : π•œ} (h : ↑βˆ₯zβˆ₯β‚Š < (spectral_radius π•œ a)⁻¹) :
is_unit (1 - z β€’ a) :=
begin
by_cases hz : z = 0,
{ simp only [hz, is_unit_one, sub_zero, zero_smul] },
{ let u := units.mk0 z hz,
suffices hu : is_unit (u⁻¹ β€’ 1 - a),
{ rwa [is_unit.smul_sub_iff_sub_inv_smul, inv_inv u] at hu },
{ rw [units.smul_def, ←algebra.algebra_map_eq_smul_one, ←mem_resolvent_set_iff],
refine mem_resolvent_set_of_spectral_radius_lt _,
rwa [units.coe_inv, nnnorm_inv, coe_inv (nnnorm_ne_zero_iff.mpr
(units.coe_mk0 hz β–Έ hz : (u : π•œ) β‰  0)), lt_inv_iff_lt_inv] } }
end
/-- In a Banach algebra `A` over `π•œ`, for `a : A` the function `Ξ» z, (1 - z β€’ a)⁻¹` is
differentiable on any closed ball centered at zero of radius `r < (spectral_radius π•œ a)⁻¹`. -/
theorem differentiable_on_inverse_one_sub_smul [complete_space A] {a : A} {r : ℝβ‰₯0}
(hr : (r : ℝβ‰₯0∞) < (spectral_radius π•œ a)⁻¹) :
differentiable_on π•œ (Ξ» z : π•œ, ring.inverse (1 - z β€’ a)) (metric.closed_ball 0 r) :=
begin
intros z z_mem,
apply differentiable_at.differentiable_within_at,
have hu : is_unit (1 - z β€’ a),
{ refine is_unit_one_sub_smul_of_lt_inv_radius (lt_of_le_of_lt (coe_mono _) hr),
simpa only [norm_to_nnreal, real.to_nnreal_coe]
using real.to_nnreal_mono (mem_closed_ball_zero_iff.mp z_mem) },
have H₁ : differentiable π•œ (Ξ» w : π•œ, 1 - w β€’ a) := (differentiable_id.smul_const a).const_sub 1,
exact differentiable_at.comp z (differentiable_at_inverse hu.unit) (H₁.differentiable_at),
end
end one_sub_smul
section gelfand_formula
open filter ennreal continuous_multilinear_map
open_locale topological_space
variables
[normed_ring A] [normed_algebra β„‚ A] [complete_space A]
/-- The `limsup` relationship for the spectral radius used to prove `spectrum.gelfand_formula`. -/
lemma limsup_pow_nnnorm_pow_one_div_le_spectral_radius (a : A) :
limsup at_top (Ξ» n : β„•, ↑βˆ₯a ^ nβˆ₯β‚Š ^ (1 / n : ℝ)) ≀ spectral_radius β„‚ a :=
begin
refine ennreal.inv_le_inv.mp (le_of_forall_pos_nnreal_lt (Ξ» r r_pos r_lt, _)),
simp_rw [inv_limsup, ←one_div],
let p : formal_multilinear_series β„‚ β„‚ A :=
Ξ» n, continuous_multilinear_map.mk_pi_field β„‚ (fin n) (a ^ n),
suffices h : (r : ℝβ‰₯0∞) ≀ p.radius,
{ convert h,
simp only [p.radius_eq_liminf, ←norm_to_nnreal, norm_mk_pi_field],
refine congr_arg _ (funext (Ξ» n, congr_arg _ _)),
rw [norm_to_nnreal, ennreal.coe_rpow_def (βˆ₯a ^ nβˆ₯β‚Š) (1 / n : ℝ), if_neg],
exact Ξ» ha, by linarith [ha.2, (one_div_nonneg.mpr n.cast_nonneg : 0 ≀ (1 / n : ℝ))], },
{ have H₁ := (differentiable_on_inverse_one_sub_smul r_lt).has_fpower_series_on_ball r_pos,
exact ((has_fpower_series_on_ball_inverse_one_sub_smul β„‚ a).exchange_radius H₁).r_le, }
end
/-- **Gelfand's formula**: Given an element `a : A` of a complex Banach algebra, the
`spectral_radius` of `a` is the limit of the sequence `βˆ₯a ^ nβˆ₯β‚Š ^ (1 / n)` -/
theorem pow_nnnorm_pow_one_div_tendsto_nhds_spectral_radius [norm_one_class A] (a : A) :
tendsto (Ξ» n : β„•, ((βˆ₯a ^ nβˆ₯β‚Š ^ (1 / n : ℝ)) : ℝβ‰₯0∞)) at_top (𝓝 (spectral_radius β„‚ a)) :=
begin
refine tendsto_of_le_liminf_of_limsup_le _ _ (by apply_auto_param) (by apply_auto_param),
{ rw [←liminf_nat_add _ 1, liminf_eq_supr_infi_of_nat],
refine le_trans _ (le_supr _ 0),
simp only [nat.cast_succ],
exact le_infiβ‚‚ (Ξ» i hi, spectral_radius_le_pow_nnnorm_pow_one_div β„‚ a i) },
{ exact limsup_pow_nnnorm_pow_one_div_le_spectral_radius a },
end
/- This is the same as `pow_nnnorm_pow_one_div_tendsto_nhds_spectral_radius` but for `norm`
instead of `nnnorm`. -/
/-- **Gelfand's formula**: Given an element `a : A` of a complex Banach algebra, the
`spectral_radius` of `a` is the limit of the sequence `βˆ₯a ^ nβˆ₯β‚Š ^ (1 / n)` -/
theorem pow_norm_pow_one_div_tendsto_nhds_spectral_radius [norm_one_class A] (a : A) :
tendsto (Ξ» n : β„•, ennreal.of_real (βˆ₯a ^ nβˆ₯ ^ (1 / n : ℝ))) at_top (𝓝 (spectral_radius β„‚ a)) :=
begin
convert pow_nnnorm_pow_one_div_tendsto_nhds_spectral_radius a,
ext1,
rw [←of_real_rpow_of_nonneg (norm_nonneg _) _, ←coe_nnnorm, coe_nnreal_eq],
exact one_div_nonneg.mpr (by exact_mod_cast zero_le _),
end
end gelfand_formula
/-- In a (nontrivial) complex Banach algebra, every element has nonempty spectrum. -/
theorem nonempty {A : Type*} [normed_ring A] [normed_algebra β„‚ A] [complete_space A]
[nontrivial A]
(a : A) : (spectrum β„‚ a).nonempty :=
begin
/- Suppose `Οƒ a = βˆ…`, then resolvent set is `β„‚`, any `(z β€’ 1 - a)` is a unit, and `resolvent`
is differentiable on `β„‚`. -/
rw ←set.ne_empty_iff_nonempty,
by_contra h,
have Hβ‚€ : resolvent_set β„‚ a = set.univ, by rwa [spectrum, set.compl_empty_iff] at h,
have H₁ : differentiable β„‚ (Ξ» z : β„‚, resolvent a z), from Ξ» z,
(has_deriv_at_resolvent (Hβ‚€.symm β–Έ set.mem_univ z : z ∈ resolvent_set β„‚ a)).differentiable_at,
/- The norm of the resolvent is small for all sufficently large `z`, and by compactness and
continuity it is bounded on the complement of a large ball, thus uniformly bounded on `β„‚`.
By Liouville's theorem `Ξ» z, resolvent a z` is constant -/
have Hβ‚‚ := norm_resolvent_le_forall a,
have H₃ : βˆ€ z : β„‚, resolvent a z = resolvent a (0 : β„‚),
{ refine Ξ» z, H₁.apply_eq_apply_of_bounded (bounded_iff_exists_norm_le.mpr _) z 0,
rcases Hβ‚‚ 1 zero_lt_one with ⟨R, R_pos, hR⟩,
rcases (proper_space.is_compact_closed_ball (0 : β„‚) R).exists_bound_of_continuous_on
H₁.continuous.continuous_on with ⟨C, hC⟩,
use max C 1,
rintros _ ⟨w, rfl⟩,
refine or.elim (em (βˆ₯wβˆ₯ ≀ R)) (Ξ» hw, _) (Ξ» hw, _),
{ exact (hC w (mem_closed_ball_zero_iff.mpr hw)).trans (le_max_left _ _) },
{ exact (hR w (not_le.mp hw).le).trans (le_max_right _ _), }, },
/- `resolvent a 0 = 0`, which is a contradition because it isn't a unit. -/
have Hβ‚… : resolvent a (0 : β„‚) = 0,
{ refine norm_eq_zero.mp (le_antisymm (le_of_forall_pos_le_add (Ξ» Ξ΅ hΞ΅, _)) (norm_nonneg _)),
rcases Hβ‚‚ Ξ΅ hΞ΅ with ⟨R, R_pos, hR⟩,
simpa only [H₃ R] using (zero_add Ξ΅).symm.subst
(hR R (by exact_mod_cast (real.norm_of_nonneg R_pos.lt.le).symm.le)), },
/- `not_is_unit_zero` is where we need `nontrivial A`, it is unavoidable. -/
exact not_is_unit_zero (Hβ‚….subst (is_unit_resolvent.mp
(mem_resolvent_set_iff.mp (Hβ‚€.symm β–Έ set.mem_univ 0)))),
end
section gelfand_mazur_isomorphism
variables [normed_division_ring A] [normed_algebra β„‚ A]
local notation `Οƒ` := spectrum β„‚
lemma algebra_map_eq_of_mem {a : A} {z : β„‚} (h : z ∈ Οƒ a) : algebra_map β„‚ A z = a :=
by rwa [mem_iff, is_unit_iff_ne_zero, not_not, sub_eq_zero] at h
/-- **Gelfand-Mazur theorem**: For a complex Banach division algebra, the natural `algebra_map β„‚ A`
is an algebra isomorphism whose inverse is given by selecting the (unique) element of
`spectrum β„‚ a`. In addition, `algebra_map_isometry` guarantees this map is an isometry. -/
@[simps]
noncomputable def _root_.normed_division_ring.alg_equiv_complex_of_complete
[complete_space A] : β„‚ ≃ₐ[β„‚] A :=
{ to_fun := algebra_map β„‚ A,
inv_fun := Ξ» a, (spectrum.nonempty a).some,
left_inv := Ξ» z, by simpa only [scalar_eq] using (spectrum.nonempty $ algebra_map β„‚ A z).some_mem,
right_inv := Ξ» a, algebra_map_eq_of_mem (spectrum.nonempty a).some_mem,
..algebra.of_id β„‚ A }
end gelfand_mazur_isomorphism
section exp_mapping
local notation `↑ₐ` := algebra_map π•œ A
/-- For `π•œ = ℝ` or `π•œ = β„‚`, `exp π•œ` maps the spectrum of `a` into the spectrum of `exp π•œ a`. -/
theorem exp_mem_exp [is_R_or_C π•œ] [normed_ring A] [normed_algebra π•œ A] [complete_space A]
(a : A) {z : π•œ} (hz : z ∈ spectrum π•œ a) : exp π•œ z ∈ spectrum π•œ (exp π•œ a) :=
begin
have hexpmul : exp π•œ a = exp π•œ (a - ↑ₐ z) * ↑ₐ (exp π•œ z),
{ rw [algebra_map_exp_comm z, ←exp_add_of_commute (algebra.commutes z (a - ↑ₐz)).symm,
sub_add_cancel] },
let b := βˆ‘' n : β„•, ((n + 1).factorial⁻¹ : π•œ) β€’ (a - ↑ₐz) ^ n,
have hb : summable (Ξ» n : β„•, ((n + 1).factorial⁻¹ : π•œ) β€’ (a - ↑ₐz) ^ n),
{ refine summable_of_norm_bounded_eventually _ (real.summable_pow_div_factorial βˆ₯a - ↑ₐzβˆ₯) _,
filter_upwards [filter.eventually_cofinite_ne 0] with n hn,
rw [norm_smul, mul_comm, norm_inv, is_R_or_C.norm_eq_abs, is_R_or_C.abs_cast_nat,
←div_eq_mul_inv],
exact div_le_div (pow_nonneg (norm_nonneg _) n) (norm_pow_le' (a - ↑ₐz) (zero_lt_iff.mpr hn))
(by exact_mod_cast nat.factorial_pos n)
(by exact_mod_cast nat.factorial_le (lt_add_one n).le) },
have hβ‚€ : βˆ‘' n : β„•, ((n + 1).factorial⁻¹ : π•œ) β€’ (a - ↑ₐz) ^ (n + 1) = (a - ↑ₐz) * b,
{ simpa only [mul_smul_comm, pow_succ] using hb.tsum_mul_left (a - ↑ₐz) },
have h₁ : βˆ‘' n : β„•, ((n + 1).factorial⁻¹ : π•œ) β€’ (a - ↑ₐz) ^ (n + 1) = b * (a - ↑ₐz),
{ simpa only [pow_succ', algebra.smul_mul_assoc] using hb.tsum_mul_right (a - ↑ₐz) },
have h₃ : exp π•œ (a - ↑ₐz) = 1 + (a - ↑ₐz) * b,
{ rw exp_eq_tsum,
convert tsum_eq_zero_add (exp_series_summable' (a - ↑ₐz)),
simp only [nat.factorial_zero, nat.cast_one, inv_one, pow_zero, one_smul],
exact hβ‚€.symm },
rw [spectrum.mem_iff, is_unit.sub_iff, ←one_mul (↑ₐ(exp π•œ z)), hexpmul, ←_root_.sub_mul,
commute.is_unit_mul_iff (algebra.commutes (exp π•œ z) (exp π•œ (a - ↑ₐz) - 1)).symm,
sub_eq_iff_eq_add'.mpr h₃, commute.is_unit_mul_iff (hβ‚€ β–Έ h₁ : (a - ↑ₐz) * b = b * (a - ↑ₐz))],
exact not_and_of_not_left _ (not_and_of_not_left _ ((not_iff_not.mpr is_unit.sub_iff).mp hz)),
end
end exp_mapping
end spectrum
namespace alg_hom
section normed_field
variables [normed_field π•œ] [normed_ring A] [normed_algebra π•œ A] [complete_space A]
local notation `↑ₐ` := algebra_map π•œ A
/-- An algebra homomorphism into the base field, as a continuous linear map (since it is
automatically bounded). -/
@[simps] def to_continuous_linear_map [norm_one_class A] (Ο† : A →ₐ[π•œ] π•œ) : A β†’L[π•œ] π•œ :=
Ο†.to_linear_map.mk_continuous_of_exists_bound $
⟨1, Ξ» a, (one_mul βˆ₯aβˆ₯).symm β–Έ spectrum.norm_le_norm_of_mem (Ο†.apply_mem_spectrum _)⟩
lemma continuous [norm_one_class A] (Ο† : A →ₐ[π•œ] π•œ) : continuous Ο† :=
Ο†.to_continuous_linear_map.continuous
end normed_field
section nontrivially_normed_field
variables [nontrivially_normed_field π•œ] [normed_ring A] [normed_algebra π•œ A] [complete_space A]
local notation `↑ₐ` := algebra_map π•œ A
@[simp] lemma to_continuous_linear_map_norm [norm_one_class A] (Ο† : A →ₐ[π•œ] π•œ) :
βˆ₯Ο†.to_continuous_linear_mapβˆ₯ = 1 :=
continuous_linear_map.op_norm_eq_of_bounds zero_le_one
(Ξ» a, (one_mul βˆ₯aβˆ₯).symm β–Έ spectrum.norm_le_norm_of_mem (Ο†.apply_mem_spectrum _))
(Ξ» _ _ h, by simpa only [to_continuous_linear_map_apply, mul_one, map_one, norm_one] using h 1)
end nontrivially_normed_field
end alg_hom