(* Title: HOL/Analysis/Summation_Tests.thy
Author: Manuel Eberl, TU München
*)
section \<open>Radius of Convergence and Summation Tests\<close>
theory Summation_Tests
imports
Complex_Main
"HOL-Library.Discrete"
"HOL-Library.Extended_Real"
"HOL-Library.Liminf_Limsup"
Extended_Real_Limits
begin
text \<open>
The definition of the radius of convergence of a power series,
various summability tests, lemmas to compute the radius of convergence etc.
\<close>
subsection \<open>Convergence tests for infinite sums\<close>
subsubsection \<open>Root test\<close>
lemma limsup_root_powser:
fixes f :: "nat \<Rightarrow> 'a :: {banach, real_normed_div_algebra}"
shows "limsup (\<lambda>n. ereal (root n (norm (f n * z ^ n)))) =
limsup (\<lambda>n. ereal (root n (norm (f n)))) * ereal (norm z)"
proof -
have A: "(\<lambda>n. ereal (root n (norm (f n * z ^ n)))) =
(\<lambda>n. ereal (root n (norm (f n))) * ereal (norm z))" (is "?g = ?h")
proof
fix n show "?g n = ?h n"
by (cases "n = 0") (simp_all add: norm_mult real_root_mult real_root_pos2 norm_power)
qed
show ?thesis by (subst A, subst limsup_ereal_mult_right) simp_all
qed
lemma limsup_root_limit:
assumes "(\<lambda>n. ereal (root n (norm (f n)))) \<longlonglongrightarrow> l" (is "?g \<longlonglongrightarrow> _")
shows "limsup (\<lambda>n. ereal (root n (norm (f n)))) = l"
proof -
from assms have "convergent ?g" "lim ?g = l"
unfolding convergent_def by (blast intro: limI)+
with convergent_limsup_cl show ?thesis by force
qed
lemma limsup_root_limit':
assumes "(\<lambda>n. root n (norm (f n))) \<longlonglongrightarrow> l"
shows "limsup (\<lambda>n. ereal (root n (norm (f n)))) = ereal l"
by (intro limsup_root_limit tendsto_ereal assms)
theorem root_test_convergence':
fixes f :: "nat \<Rightarrow> 'a :: banach"
defines "l \<equiv> limsup (\<lambda>n. ereal (root n (norm (f n))))"
assumes l: "l < 1"
shows "summable f"
proof -
have "0 = limsup (\<lambda>n. 0)" by (simp add: Limsup_const)
also have "... \<le> l" unfolding l_def by (intro Limsup_mono) (simp_all add: real_root_ge_zero)
finally have "l \<ge> 0" by simp
with l obtain l' where l': "l = ereal l'" by (cases l) simp_all
define c where "c = (1 - l') / 2"
from l and \<open>l \<ge> 0\<close> have c: "l + c > l" "l' + c \<ge> 0" "l' + c < 1" unfolding c_def
by (simp_all add: field_simps l')
have "\<forall>C>l. eventually (\<lambda>n. ereal (root n (norm (f n))) < C) sequentially"
by (subst Limsup_le_iff[symmetric]) (simp add: l_def)
with c have "eventually (\<lambda>n. ereal (root n (norm (f n))) < l + ereal c) sequentially" by simp
with eventually_gt_at_top[of "0::nat"]
have "eventually (\<lambda>n. norm (f n) \<le> (l' + c) ^ n) sequentially"
proof eventually_elim
fix n :: nat assume n: "n > 0"
assume "ereal (root n (norm (f n))) < l + ereal c"
hence "root n (norm (f n)) \<le> l' + c" by (simp add: l')
with c n have "root n (norm (f n)) ^ n \<le> (l' + c) ^ n"
by (intro power_mono) (simp_all add: real_root_ge_zero)
also from n have "root n (norm (f n)) ^ n = norm (f n)" by simp
finally show "norm (f n) \<le> (l' + c) ^ n" by simp
qed
thus ?thesis
by (rule summable_comparison_test_ev[OF _ summable_geometric]) (simp add: c)
qed
theorem root_test_divergence:
fixes f :: "nat \<Rightarrow> 'a :: banach"
defines "l \<equiv> limsup (\<lambda>n. ereal (root n (norm (f n))))"
assumes l: "l > 1"
shows "\<not>summable f"
proof
assume "summable f"
hence bounded: "Bseq f" by (simp add: summable_imp_Bseq)
have "0 = limsup (\<lambda>n. 0)" by (simp add: Limsup_const)
also have "... \<le> l" unfolding l_def by (intro Limsup_mono) (simp_all add: real_root_ge_zero)
finally have l_nonneg: "l \<ge> 0" by simp
define c where "c = (if l = \<infinity> then 2 else 1 + (real_of_ereal l - 1) / 2)"
from l l_nonneg consider "l = \<infinity>" | "\<exists>l'. l = ereal l'" by (cases l) simp_all
hence c: "c > 1 \<and> ereal c < l" by cases (insert l, auto simp: c_def field_simps)
have unbounded: "\<not>bdd_above {n. root n (norm (f n)) > c}"
proof
assume "bdd_above {n. root n (norm (f n)) > c}"
then obtain N where "\<forall>n. root n (norm (f n)) > c \<longrightarrow> n \<le> N" unfolding bdd_above_def by blast
hence "\<exists>N. \<forall>n\<ge>N. root n (norm (f n)) \<le> c"
by (intro exI[of _ "N + 1"]) (force simp: not_less_eq_eq[symmetric])
hence "eventually (\<lambda>n. root n (norm (f n)) \<le> c) sequentially"
by (auto simp: eventually_at_top_linorder)
hence "l \<le> c" unfolding l_def by (intro Limsup_bounded) simp_all
with c show False by auto
qed
from bounded obtain K where K: "K > 0" "\<And>n. norm (f n) \<le> K" using BseqE by blast
define n where "n = nat \<lceil>log c K\<rceil>"
from unbounded have "\<exists>m>n. c < root m (norm (f m))" unfolding bdd_above_def
by (auto simp: not_le)
then obtain m where m: "n < m" "c < root m (norm (f m))" by auto
from c K have "K = c powr log c K" by (simp add: powr_def log_def)
also from c have "c powr log c K \<le> c powr real n" unfolding n_def
by (intro powr_mono, linarith, simp)
finally have "K \<le> c ^ n" using c by (simp add: powr_realpow)
also from c m have "c ^ n < c ^ m" by simp
also from c m have "c ^ m < root m (norm (f m)) ^ m" by (intro power_strict_mono) simp_all
also from m have "... = norm (f m)" by simp
finally show False using K(2)[of m] by simp
qed
subsubsection \<open>Cauchy's condensation test\<close>
context
fixes f :: "nat \<Rightarrow> real"
begin
private lemma condensation_inequality:
assumes mono: "\<And>m n. 0 < m \<Longrightarrow> m \<le> n \<Longrightarrow> f n \<le> f m"
shows "(\<Sum>k=1..<n. f k) \<ge> (\<Sum>k=1..<n. f (2 * 2 ^ Discrete.log k))" (is "?thesis1")
"(\<Sum>k=1..<n. f k) \<le> (\<Sum>k=1..<n. f (2 ^ Discrete.log k))" (is "?thesis2")
by (intro sum_mono mono Discrete.log_exp2_ge Discrete.log_exp2_le, simp, simp)+
private lemma condensation_condense1: "(\<Sum>k=1..<2^n. f (2 ^ Discrete.log k)) = (\<Sum>k<n. 2^k * f (2 ^ k))"
proof (induction n)
case (Suc n)
have "{1..<2^Suc n} = {1..<2^n} \<union> {2^n..<(2^Suc n :: nat)}" by auto
also have "(\<Sum>k\<in>\<dots>. f (2 ^ Discrete.log k)) =
(\<Sum>k<n. 2^k * f (2^k)) + (\<Sum>k = 2^n..<2^Suc n. f (2^Discrete.log k))"
by (subst sum.union_disjoint) (insert Suc, auto)
also have "Discrete.log k = n" if "k \<in> {2^n..<2^Suc n}" for k using that by (intro Discrete.log_eqI) simp_all
hence "(\<Sum>k = 2^n..<2^Suc n. f (2^Discrete.log k)) = (\<Sum>(_::nat) = 2^n..<2^Suc n. f (2^n))"
by (intro sum.cong) simp_all
also have "\<dots> = 2^n * f (2^n)" by (simp)
finally show ?case by simp
qed simp
private lemma condensation_condense2: "(\<Sum>k=1..<2^n. f (2 * 2 ^ Discrete.log k)) = (\<Sum>k<n. 2^k * f (2 ^ Suc k))"
proof (induction n)
case (Suc n)
have "{1..<2^Suc n} = {1..<2^n} \<union> {2^n..<(2^Suc n :: nat)}" by auto
also have "(\<Sum>k\<in>\<dots>. f (2 * 2 ^ Discrete.log k)) =
(\<Sum>k<n. 2^k * f (2^Suc k)) + (\<Sum>k = 2^n..<2^Suc n. f (2 * 2^Discrete.log k))"
by (subst sum.union_disjoint) (insert Suc, auto)
also have "Discrete.log k = n" if "k \<in> {2^n..<2^Suc n}" for k using that by (intro Discrete.log_eqI) simp_all
hence "(\<Sum>k = 2^n..<2^Suc n. f (2*2^Discrete.log k)) = (\<Sum>(_::nat) = 2^n..<2^Suc n. f (2^Suc n))"
by (intro sum.cong) simp_all
also have "\<dots> = 2^n * f (2^Suc n)" by (simp)
finally show ?case by simp
qed simp
theorem condensation_test:
assumes mono: "\<And>m. 0 < m \<Longrightarrow> f (Suc m) \<le> f m"
assumes nonneg: "\<And>n. f n \<ge> 0"
shows "summable f \<longleftrightarrow> summable (\<lambda>n. 2^n * f (2^n))"
proof -
define f' where "f' n = (if n = 0 then 0 else f n)" for n
from mono have mono': "decseq (\<lambda>n. f (Suc n))" by (intro decseq_SucI) simp
hence mono': "f n \<le> f m" if "m \<le> n" "m > 0" for m n
using that decseqD[OF mono', of "m - 1" "n - 1"] by simp
have "(\<lambda>n. f (Suc n)) = (\<lambda>n. f' (Suc n))" by (intro ext) (simp add: f'_def)
hence "summable f \<longleftrightarrow> summable f'"
by (subst (1 2) summable_Suc_iff [symmetric]) (simp only:)
also have "\<dots> \<longleftrightarrow> convergent (\<lambda>n. \<Sum>k<n. f' k)" unfolding summable_iff_convergent ..
also have "monoseq (\<lambda>n. \<Sum>k<n. f' k)" unfolding f'_def
by (intro mono_SucI1) (auto intro!: mult_nonneg_nonneg nonneg)
hence "convergent (\<lambda>n. \<Sum>k<n. f' k) \<longleftrightarrow> Bseq (\<lambda>n. \<Sum>k<n. f' k)"
by (rule monoseq_imp_convergent_iff_Bseq)
also have "\<dots> \<longleftrightarrow> Bseq (\<lambda>n. \<Sum>k=1..<n. f' k)" unfolding One_nat_def
by (subst sum_shift_lb_Suc0_0_upt) (simp_all add: f'_def atLeast0LessThan)
also have "\<dots> \<longleftrightarrow> Bseq (\<lambda>n. \<Sum>k=1..<n. f k)" unfolding f'_def by simp
also have "\<dots> \<longleftrightarrow> Bseq (\<lambda>n. \<Sum>k=1..<2^n. f k)"
by (rule nonneg_incseq_Bseq_subseq_iff[symmetric])
(auto intro!: sum_nonneg incseq_SucI nonneg simp: strict_mono_def)
also have "\<dots> \<longleftrightarrow> Bseq (\<lambda>n. \<Sum>k<n. 2^k * f (2^k))"
proof (intro iffI)
assume A: "Bseq (\<lambda>n. \<Sum>k=1..<2^n. f k)"
have "eventually (\<lambda>n. norm (\<Sum>k<n. 2^k * f (2^Suc k)) \<le> norm (\<Sum>k=1..<2^n. f k)) sequentially"
proof (intro always_eventually allI)
fix n :: nat
have "norm (\<Sum>k<n. 2^k * f (2^Suc k)) = (\<Sum>k<n. 2^k * f (2^Suc k))" unfolding real_norm_def
by (intro abs_of_nonneg sum_nonneg ballI mult_nonneg_nonneg nonneg) simp_all
also have "\<dots> \<le> (\<Sum>k=1..<2^n. f k)"
by (subst condensation_condense2 [symmetric]) (intro condensation_inequality mono')
also have "\<dots> = norm \<dots>" unfolding real_norm_def
by (intro abs_of_nonneg[symmetric] sum_nonneg ballI mult_nonneg_nonneg nonneg)
finally show "norm (\<Sum>k<n. 2 ^ k * f (2 ^ Suc k)) \<le> norm (\<Sum>k=1..<2^n. f k)" .
qed
from this and A have "Bseq (\<lambda>n. \<Sum>k<n. 2^k * f (2^Suc k))" by (rule Bseq_eventually_mono)
from Bseq_mult[OF Bfun_const[of 2] this] have "Bseq (\<lambda>n. \<Sum>k<n. 2^Suc k * f (2^Suc k))"
by (simp add: sum_distrib_left sum_distrib_right mult_ac)
hence "Bseq (\<lambda>n. (\<Sum>k=Suc 0..<Suc n. 2^k * f (2^k)) + f 1)"
by (intro Bseq_add, subst sum.shift_bounds_Suc_ivl) (simp add: atLeast0LessThan)
hence "Bseq (\<lambda>n. (\<Sum>k=0..<Suc n. 2^k * f (2^k)))"
by (subst sum.atLeast_Suc_lessThan) (simp_all add: add_ac)
thus "Bseq (\<lambda>n. (\<Sum>k<n. 2^k * f (2^k)))"
by (subst (asm) Bseq_Suc_iff) (simp add: atLeast0LessThan)
next
assume A: "Bseq (\<lambda>n. (\<Sum>k<n. 2^k * f (2^k)))"
have "eventually (\<lambda>n. norm (\<Sum>k=1..<2^n. f k) \<le> norm (\<Sum>k<n. 2^k * f (2^k))) sequentially"
proof (intro always_eventually allI)
fix n :: nat
have "norm (\<Sum>k=1..<2^n. f k) = (\<Sum>k=1..<2^n. f k)" unfolding real_norm_def
by (intro abs_of_nonneg sum_nonneg ballI mult_nonneg_nonneg nonneg)
also have "\<dots> \<le> (\<Sum>k<n. 2^k * f (2^k))"
by (subst condensation_condense1 [symmetric]) (intro condensation_inequality mono')
also have "\<dots> = norm \<dots>" unfolding real_norm_def
by (intro abs_of_nonneg [symmetric] sum_nonneg ballI mult_nonneg_nonneg nonneg) simp_all
finally show "norm (\<Sum>k=1..<2^n. f k) \<le> norm (\<Sum>k<n. 2^k * f (2^k))" .
qed
from this and A show "Bseq (\<lambda>n. \<Sum>k=1..<2^n. f k)" by (rule Bseq_eventually_mono)
qed
also have "monoseq (\<lambda>n. (\<Sum>k<n. 2^k * f (2^k)))"
by (intro mono_SucI1) (auto intro!: mult_nonneg_nonneg nonneg)
hence "Bseq (\<lambda>n. (\<Sum>k<n. 2^k * f (2^k))) \<longleftrightarrow> convergent (\<lambda>n. (\<Sum>k<n. 2^k * f (2^k)))"
by (rule monoseq_imp_convergent_iff_Bseq [symmetric])
also have "\<dots> \<longleftrightarrow> summable (\<lambda>k. 2^k * f (2^k))" by (simp only: summable_iff_convergent)
finally show ?thesis .
qed
end
subsubsection \<open>Summability of powers\<close>
lemma abs_summable_complex_powr_iff:
"summable (\<lambda>n. norm (exp (of_real (ln (of_nat n)) * s))) \<longleftrightarrow> Re s < -1"
proof (cases "Re s \<le> 0")
let ?l = "\<lambda>n. complex_of_real (ln (of_nat n))"
case False
have "eventually (\<lambda>n. norm (1 :: real) \<le> norm (exp (?l n * s))) sequentially"
apply (rule eventually_mono [OF eventually_gt_at_top[of "0::nat"]])
using False ge_one_powr_ge_zero by auto
from summable_comparison_test_ev[OF this] False show ?thesis by (auto simp: summable_const_iff)
next
let ?l = "\<lambda>n. complex_of_real (ln (of_nat n))"
case True
hence "summable (\<lambda>n. norm (exp (?l n * s))) \<longleftrightarrow> summable (\<lambda>n. 2^n * norm (exp (?l (2^n) * s)))"
by (intro condensation_test) (auto intro!: mult_right_mono_neg)
also have "(\<lambda>n. 2^n * norm (exp (?l (2^n) * s))) = (\<lambda>n. (2 powr (Re s + 1)) ^ n)"
proof
fix n :: nat
have "2^n * norm (exp (?l (2^n) * s)) = exp (real n * ln 2) * exp (real n * ln 2 * Re s)"
using True by (subst exp_of_nat_mult) (simp add: ln_realpow algebra_simps)
also have "\<dots> = exp (real n * (ln 2 * (Re s + 1)))"
by (simp add: algebra_simps exp_add)
also have "\<dots> = exp (ln 2 * (Re s + 1)) ^ n" by (subst exp_of_nat_mult) simp
also have "exp (ln 2 * (Re s + 1)) = 2 powr (Re s + 1)" by (simp add: powr_def)
finally show "2^n * norm (exp (?l (2^n) * s)) = (2 powr (Re s + 1)) ^ n" .
qed
also have "summable \<dots> \<longleftrightarrow> 2 powr (Re s + 1) < 2 powr 0"
by (subst summable_geometric_iff) simp
also have "\<dots> \<longleftrightarrow> Re s < -1" by (subst powr_less_cancel_iff) (simp, linarith)
finally show ?thesis .
qed
theorem summable_complex_powr_iff:
assumes "Re s < -1"
shows "summable (\<lambda>n. exp (of_real (ln (of_nat n)) * s))"
by (rule summable_norm_cancel, subst abs_summable_complex_powr_iff) fact
lemma summable_real_powr_iff: "summable (\<lambda>n. of_nat n powr s :: real) \<longleftrightarrow> s < -1"
proof -
from eventually_gt_at_top[of "0::nat"]
have "summable (\<lambda>n. of_nat n powr s) \<longleftrightarrow> summable (\<lambda>n. exp (ln (of_nat n) * s))"
by (intro summable_cong) (auto elim!: eventually_mono simp: powr_def)
also have "\<dots> \<longleftrightarrow> s < -1" using abs_summable_complex_powr_iff[of "of_real s"] by simp
finally show ?thesis .
qed
lemma inverse_power_summable:
assumes s: "s \<ge> 2"
shows "summable (\<lambda>n. inverse (of_nat n ^ s :: 'a :: {real_normed_div_algebra,banach}))"
proof (rule summable_norm_cancel, subst summable_cong)
from eventually_gt_at_top[of "0::nat"]
show "eventually (\<lambda>n. norm (inverse (of_nat n ^ s:: 'a)) = real_of_nat n powr (-real s)) at_top"
by eventually_elim (simp add: norm_inverse norm_power powr_minus powr_realpow)
qed (insert s summable_real_powr_iff[of "-s"], simp_all)
lemma not_summable_harmonic: "\<not>summable (\<lambda>n. inverse (of_nat n) :: 'a :: real_normed_field)"
proof
assume "summable (\<lambda>n. inverse (of_nat n) :: 'a)"
hence "convergent (\<lambda>n. norm (of_real (\<Sum>k<n. inverse (of_nat k)) :: 'a))"
by (simp add: summable_iff_convergent convergent_norm)
hence "convergent (\<lambda>n. abs (\<Sum>k<n. inverse (of_nat k)) :: real)" by (simp only: norm_of_real)
also have "(\<lambda>n. abs (\<Sum>k<n. inverse (of_nat k)) :: real) = (\<lambda>n. \<Sum>k<n. inverse (of_nat k))"
by (intro ext abs_of_nonneg sum_nonneg) auto
also have "convergent \<dots> \<longleftrightarrow> summable (\<lambda>k. inverse (of_nat k) :: real)"
by (simp add: summable_iff_convergent)
finally show False using summable_real_powr_iff[of "-1"] by (simp add: powr_minus)
qed
subsubsection \<open>Kummer's test\<close>
theorem kummers_test_convergence:
fixes f p :: "nat \<Rightarrow> real"
assumes pos_f: "eventually (\<lambda>n. f n > 0) sequentially"
assumes nonneg_p: "eventually (\<lambda>n. p n \<ge> 0) sequentially"
defines "l \<equiv> liminf (\<lambda>n. ereal (p n * f n / f (Suc n) - p (Suc n)))"
assumes l: "l > 0"
shows "summable f"
unfolding summable_iff_convergent'
proof -
define r where "r = (if l = \<infinity> then 1 else real_of_ereal l / 2)"
from l have "r > 0 \<and> of_real r < l" by (cases l) (simp_all add: r_def)
hence r: "r > 0" "of_real r < l" by simp_all
hence "eventually (\<lambda>n. p n * f n / f (Suc n) - p (Suc n) > r) sequentially"
unfolding l_def by (force dest: less_LiminfD)
moreover from pos_f have "eventually (\<lambda>n. f (Suc n) > 0) sequentially"
by (subst eventually_sequentially_Suc)
ultimately have "eventually (\<lambda>n. p n * f n - p (Suc n) * f (Suc n) > r * f (Suc n)) sequentially"
by eventually_elim (simp add: field_simps)
from eventually_conj[OF pos_f eventually_conj[OF nonneg_p this]]
obtain m where m: "\<And>n. n \<ge> m \<Longrightarrow> f n > 0" "\<And>n. n \<ge> m \<Longrightarrow> p n \<ge> 0"
"\<And>n. n \<ge> m \<Longrightarrow> p n * f n - p (Suc n) * f (Suc n) > r * f (Suc n)"
unfolding eventually_at_top_linorder by blast
let ?c = "(norm (\<Sum>k\<le>m. r * f k) + p m * f m) / r"
have "Bseq (\<lambda>n. (\<Sum>k\<le>n + Suc m. f k))"
proof (rule BseqI')
fix k :: nat
define n where "n = k + Suc m"
have n: "n > m" by (simp add: n_def)
from r have "r * norm (\<Sum>k\<le>n. f k) = norm (\<Sum>k\<le>n. r * f k)"
by (simp add: sum_distrib_left[symmetric] abs_mult)
also from n have "{..n} = {..m} \<union> {Suc m..n}" by auto
hence "(\<Sum>k\<le>n. r * f k) = (\<Sum>k\<in>{..m} \<union> {Suc m..n}. r * f k)" by (simp only:)
also have "\<dots> = (\<Sum>k\<le>m. r * f k) + (\<Sum>k=Suc m..n. r * f k)"
by (subst sum.union_disjoint) auto
also have "norm \<dots> \<le> norm (\<Sum>k\<le>m. r * f k) + norm (\<Sum>k=Suc m..n. r * f k)"
by (rule norm_triangle_ineq)
also from r less_imp_le[OF m(1)] have "(\<Sum>k=Suc m..n. r * f k) \<ge> 0"
by (intro sum_nonneg) auto
hence "norm (\<Sum>k=Suc m..n. r * f k) = (\<Sum>k=Suc m..n. r * f k)" by simp
also have "(\<Sum>k=Suc m..n. r * f k) = (\<Sum>k=m..<n. r * f (Suc k))"
by (subst sum.shift_bounds_Suc_ivl [symmetric])
(simp only: atLeastLessThanSuc_atLeastAtMost)
also from m have "\<dots> \<le> (\<Sum>k=m..<n. p k * f k - p (Suc k) * f (Suc k))"
by (intro sum_mono[OF less_imp_le]) simp_all
also have "\<dots> = -(\<Sum>k=m..<n. p (Suc k) * f (Suc k) - p k * f k)"
by (simp add: sum_negf [symmetric] algebra_simps)
also from n have "\<dots> = p m * f m - p n * f n"
by (cases n, simp, simp only: atLeastLessThanSuc_atLeastAtMost, subst sum_Suc_diff) simp_all
also from less_imp_le[OF m(1)] m(2) n have "\<dots> \<le> p m * f m" by simp
finally show "norm (\<Sum>k\<le>n. f k) \<le> (norm (\<Sum>k\<le>m. r * f k) + p m * f m) / r" using r
by (subst pos_le_divide_eq[OF r(1)]) (simp only: mult_ac)
qed
moreover have "(\<Sum>k\<le>n. f k) \<le> (\<Sum>k\<le>n'. f k)" if "Suc m \<le> n" "n \<le> n'" for n n'
using less_imp_le[OF m(1)] that by (intro sum_mono2) auto
ultimately show "convergent (\<lambda>n. \<Sum>k\<le>n. f k)" by (rule Bseq_monoseq_convergent'_inc)
qed
theorem kummers_test_divergence:
fixes f p :: "nat \<Rightarrow> real"
assumes pos_f: "eventually (\<lambda>n. f n > 0) sequentially"
assumes pos_p: "eventually (\<lambda>n. p n > 0) sequentially"
assumes divergent_p: "\<not>summable (\<lambda>n. inverse (p n))"
defines "l \<equiv> limsup (\<lambda>n. ereal (p n * f n / f (Suc n) - p (Suc n)))"
assumes l: "l < 0"
shows "\<not>summable f"
proof
assume "summable f"
from eventually_conj[OF pos_f eventually_conj[OF pos_p Limsup_lessD[OF l[unfolded l_def]]]]
obtain N where N: "\<And>n. n \<ge> N \<Longrightarrow> p n > 0" "\<And>n. n \<ge> N \<Longrightarrow> f n > 0"
"\<And>n. n \<ge> N \<Longrightarrow> p n * f n / f (Suc n) - p (Suc n) < 0"
by (auto simp: eventually_at_top_linorder)
hence A: "p n * f n < p (Suc n) * f (Suc n)" if "n \<ge> N" for n using that N[of n] N[of "Suc n"]
by (simp add: field_simps)
have B: "p n * f n \<ge> p N * f N" if "n \<ge> N" for n using that and A
by (induction n rule: dec_induct) (auto intro!: less_imp_le elim!: order.trans)
have "eventually (\<lambda>n. norm (p N * f N * inverse (p n)) \<le> f n) sequentially"
apply (rule eventually_mono [OF eventually_ge_at_top[of N]])
using B N by (auto simp: field_simps abs_of_pos)
from this and \<open>summable f\<close> have "summable (\<lambda>n. p N * f N * inverse (p n))"
by (rule summable_comparison_test_ev)
from summable_mult[OF this, of "inverse (p N * f N)"] N[OF le_refl]
have "summable (\<lambda>n. inverse (p n))" by (simp add: field_split_simps)
with divergent_p show False by contradiction
qed
subsubsection \<open>Ratio test\<close>
theorem ratio_test_convergence:
fixes f :: "nat \<Rightarrow> real"
assumes pos_f: "eventually (\<lambda>n. f n > 0) sequentially"
defines "l \<equiv> liminf (\<lambda>n. ereal (f n / f (Suc n)))"
assumes l: "l > 1"
shows "summable f"
proof (rule kummers_test_convergence[OF pos_f])
note l
also have "l = liminf (\<lambda>n. ereal (f n / f (Suc n) - 1)) + 1"
by (subst Liminf_add_ereal_right[symmetric]) (simp_all add: minus_ereal_def l_def one_ereal_def)
finally show "liminf (\<lambda>n. ereal (1 * f n / f (Suc n) - 1)) > 0"
by (cases "liminf (\<lambda>n. ereal (1 * f n / f (Suc n) - 1))") simp_all
qed simp
theorem ratio_test_divergence:
fixes f :: "nat \<Rightarrow> real"
assumes pos_f: "eventually (\<lambda>n. f n > 0) sequentially"
defines "l \<equiv> limsup (\<lambda>n. ereal (f n / f (Suc n)))"
assumes l: "l < 1"
shows "\<not>summable f"
proof (rule kummers_test_divergence[OF pos_f])
have "limsup (\<lambda>n. ereal (f n / f (Suc n) - 1)) + 1 = l"
by (subst Limsup_add_ereal_right[symmetric]) (simp_all add: minus_ereal_def l_def one_ereal_def)
also note l
finally show "limsup (\<lambda>n. ereal (1 * f n / f (Suc n) - 1)) < 0"
by (cases "limsup (\<lambda>n. ereal (1 * f n / f (Suc n) - 1))") simp_all
qed (simp_all add: summable_const_iff)
subsubsection \<open>Raabe's test\<close>
theorem raabes_test_convergence:
fixes f :: "nat \<Rightarrow> real"
assumes pos: "eventually (\<lambda>n. f n > 0) sequentially"
defines "l \<equiv> liminf (\<lambda>n. ereal (of_nat n * (f n / f (Suc n) - 1)))"
assumes l: "l > 1"
shows "summable f"
proof (rule kummers_test_convergence)
let ?l' = "liminf (\<lambda>n. ereal (of_nat n * f n / f (Suc n) - of_nat (Suc n)))"
have "1 < l" by fact
also have "l = liminf (\<lambda>n. ereal (of_nat n * f n / f (Suc n) - of_nat (Suc n)) + 1)"
by (simp add: l_def algebra_simps)
also have "\<dots> = ?l' + 1" by (subst Liminf_add_ereal_right) simp_all
finally show "?l' > 0" by (cases ?l') (simp_all add: algebra_simps)
qed (simp_all add: pos)
theorem raabes_test_divergence:
fixes f :: "nat \<Rightarrow> real"
assumes pos: "eventually (\<lambda>n. f n > 0) sequentially"
defines "l \<equiv> limsup (\<lambda>n. ereal (of_nat n * (f n / f (Suc n) - 1)))"
assumes l: "l < 1"
shows "\<not>summable f"
proof (rule kummers_test_divergence)
let ?l' = "limsup (\<lambda>n. ereal (of_nat n * f n / f (Suc n) - of_nat (Suc n)))"
note l
also have "l = limsup (\<lambda>n. ereal (of_nat n * f n / f (Suc n) - of_nat (Suc n)) + 1)"
by (simp add: l_def algebra_simps)
also have "\<dots> = ?l' + 1" by (subst Limsup_add_ereal_right) simp_all
finally show "?l' < 0" by (cases ?l') (simp_all add: algebra_simps)
qed (insert pos eventually_gt_at_top[of "0::nat"] not_summable_harmonic, simp_all)
subsection \<open>Radius of convergence\<close>
text \<open>
The radius of convergence of a power series. This value always exists, ranges from
\<^term>\<open>0::ereal\<close> to \<^term>\<open>\<infinity>::ereal\<close>, and the power series is guaranteed to converge for
all inputs with a norm that is smaller than that radius and to diverge for all inputs with a
norm that is greater.
\<close>
definition\<^marker>\<open>tag important\<close> conv_radius :: "(nat \<Rightarrow> 'a :: banach) \<Rightarrow> ereal" where
"conv_radius f = inverse (limsup (\<lambda>n. ereal (root n (norm (f n)))))"
lemma conv_radius_cong_weak [cong]: "(\<And>n. f n = g n) \<Longrightarrow> conv_radius f = conv_radius g"
by (drule ext) simp_all
lemma conv_radius_nonneg: "conv_radius f \<ge> 0"
proof -
have "0 = limsup (\<lambda>n. 0)" by (subst Limsup_const) simp_all
also have "\<dots> \<le> limsup (\<lambda>n. ereal (root n (norm (f n))))"
by (intro Limsup_mono) (simp_all add: real_root_ge_zero)
finally show ?thesis
unfolding conv_radius_def by (auto simp: ereal_inverse_nonneg_iff)
qed
lemma conv_radius_zero [simp]: "conv_radius (\<lambda>_. 0) = \<infinity>"
by (auto simp: conv_radius_def zero_ereal_def [symmetric] Limsup_const)
lemma conv_radius_altdef:
"conv_radius f = liminf (\<lambda>n. inverse (ereal (root n (norm (f n)))))"
by (subst Liminf_inverse_ereal) (simp_all add: real_root_ge_zero conv_radius_def)
lemma conv_radius_cong':
assumes "eventually (\<lambda>x. f x = g x) sequentially"
shows "conv_radius f = conv_radius g"
unfolding conv_radius_altdef by (intro Liminf_eq eventually_mono [OF assms]) auto
lemma conv_radius_cong:
assumes "eventually (\<lambda>x. norm (f x) = norm (g x)) sequentially"
shows "conv_radius f = conv_radius g"
unfolding conv_radius_altdef by (intro Liminf_eq eventually_mono [OF assms]) auto
theorem abs_summable_in_conv_radius:
fixes f :: "nat \<Rightarrow> 'a :: {banach, real_normed_div_algebra}"
assumes "ereal (norm z) < conv_radius f"
shows "summable (\<lambda>n. norm (f n * z ^ n))"
proof (rule root_test_convergence')
define l where "l = limsup (\<lambda>n. ereal (root n (norm (f n))))"
have "0 = limsup (\<lambda>n. 0)" by (simp add: Limsup_const)
also have "... \<le> l" unfolding l_def by (intro Limsup_mono) (simp_all add: real_root_ge_zero)
finally have l_nonneg: "l \<ge> 0" .
have "limsup (\<lambda>n. root n (norm (f n * z^n))) = l * ereal (norm z)" unfolding l_def
by (rule limsup_root_powser)
also from l_nonneg consider "l = 0" | "l = \<infinity>" | "\<exists>l'. l = ereal l' \<and> l' > 0"
by (cases "l") (auto simp: less_le)
hence "l * ereal (norm z) < 1"
proof cases
assume "l = \<infinity>"
hence "conv_radius f = 0" unfolding conv_radius_def l_def by simp
with assms show ?thesis by simp
next
assume "\<exists>l'. l = ereal l' \<and> l' > 0"
then obtain l' where l': "l = ereal l'" "0 < l'" by auto
hence "l \<noteq> \<infinity>" by auto
have "l * ereal (norm z) < l * conv_radius f"
by (intro ereal_mult_strict_left_mono) (simp_all add: l' assms)
also have "conv_radius f = inverse l" by (simp add: conv_radius_def l_def)
also from l' have "l * inverse l = 1" by simp
finally show ?thesis .
qed simp_all
finally show "limsup (\<lambda>n. ereal (root n (norm (norm (f n * z ^ n))))) < 1" by simp
qed
lemma summable_in_conv_radius:
fixes f :: "nat \<Rightarrow> 'a :: {banach, real_normed_div_algebra}"
assumes "ereal (norm z) < conv_radius f"
shows "summable (\<lambda>n. f n * z ^ n)"
by (rule summable_norm_cancel, rule abs_summable_in_conv_radius) fact+
theorem not_summable_outside_conv_radius:
fixes f :: "nat \<Rightarrow> 'a :: {banach, real_normed_div_algebra}"
assumes "ereal (norm z) > conv_radius f"
shows "\<not>summable (\<lambda>n. f n * z ^ n)"
proof (rule root_test_divergence)
define l where "l = limsup (\<lambda>n. ereal (root n (norm (f n))))"
have "0 = limsup (\<lambda>n. 0)" by (simp add: Limsup_const)
also have "... \<le> l" unfolding l_def by (intro Limsup_mono) (simp_all add: real_root_ge_zero)
finally have l_nonneg: "l \<ge> 0" .
from assms have l_nz: "l \<noteq> 0" unfolding conv_radius_def l_def by auto
have "limsup (\<lambda>n. ereal (root n (norm (f n * z^n)))) = l * ereal (norm z)"
unfolding l_def by (rule limsup_root_powser)
also have "... > 1"
proof (cases l)
assume "l = \<infinity>"
with assms conv_radius_nonneg[of f] show ?thesis
by (auto simp: zero_ereal_def[symmetric])
next
fix l' assume l': "l = ereal l'"
from l_nonneg l_nz have "1 = l * inverse l" by (auto simp: l' field_simps)
also from l_nz have "inverse l = conv_radius f"
unfolding l_def conv_radius_def by auto
also from l' l_nz l_nonneg assms have "l * \<dots> < l * ereal (norm z)"
by (intro ereal_mult_strict_left_mono) (auto simp: l')
finally show ?thesis .
qed (insert l_nonneg, simp_all)
finally show "limsup (\<lambda>n. ereal (root n (norm (f n * z^n)))) > 1" .
qed
lemma conv_radius_geI:
assumes "summable (\<lambda>n. f n * z ^ n :: 'a :: {banach, real_normed_div_algebra})"
shows "conv_radius f \<ge> norm z"
using not_summable_outside_conv_radius[of f z] assms by (force simp: not_le[symmetric])
lemma conv_radius_leI:
assumes "\<not>summable (\<lambda>n. norm (f n * z ^ n :: 'a :: {banach, real_normed_div_algebra}))"
shows "conv_radius f \<le> norm z"
using abs_summable_in_conv_radius[of z f] assms by (force simp: not_le[symmetric])
lemma conv_radius_leI':
assumes "\<not>summable (\<lambda>n. f n * z ^ n :: 'a :: {banach, real_normed_div_algebra})"
shows "conv_radius f \<le> norm z"
using summable_in_conv_radius[of z f] assms by (force simp: not_le[symmetric])
lemma conv_radius_geI_ex:
fixes f :: "nat \<Rightarrow> 'a :: {banach, real_normed_div_algebra}"
assumes "\<And>r. 0 < r \<Longrightarrow> ereal r < R \<Longrightarrow> \<exists>z. norm z = r \<and> summable (\<lambda>n. f n * z^n)"
shows "conv_radius f \<ge> R"
proof (rule linorder_cases[of "conv_radius f" R])
assume R: "conv_radius f < R"
with conv_radius_nonneg[of f] obtain conv_radius'
where [simp]: "conv_radius f = ereal conv_radius'"
by (cases "conv_radius f") simp_all
define r where "r = (if R = \<infinity> then conv_radius' + 1 else (real_of_ereal R + conv_radius') / 2)"
from R conv_radius_nonneg[of f] have "0 < r \<and> ereal r < R \<and> ereal r > conv_radius f"
unfolding r_def by (cases R) (auto simp: r_def field_simps)
with assms(1)[of r] obtain z where "norm z > conv_radius f" "summable (\<lambda>n. f n * z^n)" by auto
with not_summable_outside_conv_radius[of f z] show ?thesis by simp
qed simp_all
lemma conv_radius_geI_ex':
fixes f :: "nat \<Rightarrow> 'a :: {banach, real_normed_div_algebra}"
assumes "\<And>r. 0 < r \<Longrightarrow> ereal r < R \<Longrightarrow> summable (\<lambda>n. f n * of_real r^n)"
shows "conv_radius f \<ge> R"
proof (rule conv_radius_geI_ex)
fix r assume "0 < r" "ereal r < R"
with assms[of r] show "\<exists>z. norm z = r \<and> summable (\<lambda>n. f n * z ^ n)"
by (intro exI[of _ "of_real r :: 'a"]) auto
qed
lemma conv_radius_leI_ex:
fixes f :: "nat \<Rightarrow> 'a :: {banach, real_normed_div_algebra}"
assumes "R \<ge> 0"
assumes "\<And>r. 0 < r \<Longrightarrow> ereal r > R \<Longrightarrow> \<exists>z. norm z = r \<and> \<not>summable (\<lambda>n. norm (f n * z^n))"
shows "conv_radius f \<le> R"
proof (rule linorder_cases[of "conv_radius f" R])
assume R: "conv_radius f > R"
from R assms(1) obtain R' where R': "R = ereal R'" by (cases R) simp_all
define r where
"r = (if conv_radius f = \<infinity> then R' + 1 else (R' + real_of_ereal (conv_radius f)) / 2)"
from R conv_radius_nonneg[of f] have "r > R \<and> r < conv_radius f" unfolding r_def
by (cases "conv_radius f") (auto simp: r_def field_simps R')
with assms(1) assms(2)[of r] R'
obtain z where "norm z < conv_radius f" "\<not>summable (\<lambda>n. norm (f n * z^n))" by auto
with abs_summable_in_conv_radius[of z f] show ?thesis by auto
qed simp_all
lemma conv_radius_leI_ex':
fixes f :: "nat \<Rightarrow> 'a :: {banach, real_normed_div_algebra}"
assumes "R \<ge> 0"
assumes "\<And>r. 0 < r \<Longrightarrow> ereal r > R \<Longrightarrow> \<not>summable (\<lambda>n. f n * of_real r^n)"
shows "conv_radius f \<le> R"
proof (rule conv_radius_leI_ex)
fix r assume "0 < r" "ereal r > R"
with assms(2)[of r] show "\<exists>z. norm z = r \<and> \<not>summable (\<lambda>n. norm (f n * z ^ n))"
by (intro exI[of _ "of_real r :: 'a"]) (auto dest: summable_norm_cancel)
qed fact+
lemma conv_radius_eqI:
fixes f :: "nat \<Rightarrow> 'a :: {banach, real_normed_div_algebra}"
assumes "R \<ge> 0"
assumes "\<And>r. 0 < r \<Longrightarrow> ereal r < R \<Longrightarrow> \<exists>z. norm z = r \<and> summable (\<lambda>n. f n * z^n)"
assumes "\<And>r. 0 < r \<Longrightarrow> ereal r > R \<Longrightarrow> \<exists>z. norm z = r \<and> \<not>summable (\<lambda>n. norm (f n * z^n))"
shows "conv_radius f = R"
by (intro antisym conv_radius_geI_ex conv_radius_leI_ex assms)
lemma conv_radius_eqI':
fixes f :: "nat \<Rightarrow> 'a :: {banach, real_normed_div_algebra}"
assumes "R \<ge> 0"
assumes "\<And>r. 0 < r \<Longrightarrow> ereal r < R \<Longrightarrow> summable (\<lambda>n. f n * (of_real r)^n)"
assumes "\<And>r. 0 < r \<Longrightarrow> ereal r > R \<Longrightarrow> \<not>summable (\<lambda>n. norm (f n * (of_real r)^n))"
shows "conv_radius f = R"
proof (intro conv_radius_eqI[OF assms(1)])
fix r assume "0 < r" "ereal r < R" with assms(2)[OF this]
show "\<exists>z. norm z = r \<and> summable (\<lambda>n. f n * z ^ n)" by force
next
fix r assume "0 < r" "ereal r > R" with assms(3)[OF this]
show "\<exists>z. norm z = r \<and> \<not>summable (\<lambda>n. norm (f n * z ^ n))" by force
qed
lemma conv_radius_zeroI:
fixes f :: "nat \<Rightarrow> 'a :: {banach,real_normed_div_algebra}"
assumes "\<And>z. z \<noteq> 0 \<Longrightarrow> \<not>summable (\<lambda>n. f n * z^n)"
shows "conv_radius f = 0"
proof (rule ccontr)
assume "conv_radius f \<noteq> 0"
with conv_radius_nonneg[of f] have pos: "conv_radius f > 0" by simp
define r where "r = (if conv_radius f = \<infinity> then 1 else real_of_ereal (conv_radius f) / 2)"
from pos have r: "ereal r > 0 \<and> ereal r < conv_radius f"
by (cases "conv_radius f") (simp_all add: r_def)
hence "summable (\<lambda>n. f n * of_real r ^ n)" by (intro summable_in_conv_radius) simp
moreover from r and assms[of "of_real r"] have "\<not>summable (\<lambda>n. f n * of_real r ^ n)" by simp
ultimately show False by contradiction
qed
lemma conv_radius_inftyI':
fixes f :: "nat \<Rightarrow> 'a :: {banach,real_normed_div_algebra}"
assumes "\<And>r. r > c \<Longrightarrow> \<exists>z. norm z = r \<and> summable (\<lambda>n. f n * z^n)"
shows "conv_radius f = \<infinity>"
proof -
{
fix r :: real
have "max r (c + 1) > c" by (auto simp: max_def)
from assms[OF this] obtain z where "norm z = max r (c + 1)" "summable (\<lambda>n. f n * z^n)" by blast
from conv_radius_geI[OF this(2)] this(1) have "conv_radius f \<ge> r" by simp
}
from this[of "real_of_ereal (conv_radius f + 1)"] show "conv_radius f = \<infinity>"
by (cases "conv_radius f") simp_all
qed
lemma conv_radius_inftyI:
fixes f :: "nat \<Rightarrow> 'a :: {banach,real_normed_div_algebra}"
assumes "\<And>r. \<exists>z. norm z = r \<and> summable (\<lambda>n. f n * z^n)"
shows "conv_radius f = \<infinity>"
using assms by (rule conv_radius_inftyI')
lemma conv_radius_inftyI'':
fixes f :: "nat \<Rightarrow> 'a :: {banach,real_normed_div_algebra}"
assumes "\<And>z. summable (\<lambda>n. f n * z^n)"
shows "conv_radius f = \<infinity>"
proof (rule conv_radius_inftyI')
fix r :: real assume "r > 0"
with assms show "\<exists>z. norm z = r \<and> summable (\<lambda>n. f n * z^n)"
by (intro exI[of _ "of_real r"]) simp
qed
lemma conv_radius_conv_Sup:
fixes f :: "nat \<Rightarrow> 'a :: {banach, real_normed_div_algebra}"
shows "conv_radius f = Sup {r. \<forall>z. ereal (norm z) < r \<longrightarrow> summable (\<lambda>n. f n * z ^ n)}"
proof (rule Sup_eqI [symmetric], goal_cases)
case (1 r)
thus ?case
by (intro conv_radius_geI_ex') auto
next
case (2 r)
from 2[of 0] have r: "r \<ge> 0" by auto
show ?case
proof (intro conv_radius_leI_ex' r)
fix R assume R: "R > 0" "ereal R > r"
with r obtain r' where [simp]: "r = ereal r'" by (cases r) auto
show "\<not>summable (\<lambda>n. f n * of_real R ^ n)"
proof
assume *: "summable (\<lambda>n. f n * of_real R ^ n)"
define R' where "R' = (R + r') / 2"
from R have R': "R' > r'" "R' < R" by (simp_all add: R'_def)
hence "\<forall>z. norm z < R' \<longrightarrow> summable (\<lambda>n. f n * z ^ n)"
using powser_inside[OF *] by auto
from 2[of R'] and this have "R' \<le> r'" by auto
with \<open>R' > r'\<close> show False by simp
qed
qed
qed
lemma conv_radius_shift:
fixes f :: "nat \<Rightarrow> 'a :: {banach, real_normed_div_algebra}"
shows "conv_radius (\<lambda>n. f (n + m)) = conv_radius f"
unfolding conv_radius_conv_Sup summable_powser_ignore_initial_segment ..
lemma conv_radius_norm [simp]: "conv_radius (\<lambda>x. norm (f x)) = conv_radius f"
by (simp add: conv_radius_def)
lemma conv_radius_ratio_limit_ereal:
fixes f :: "nat \<Rightarrow> 'a :: {banach,real_normed_div_algebra}"
assumes nz: "eventually (\<lambda>n. f n \<noteq> 0) sequentially"
assumes lim: "(\<lambda>n. ereal (norm (f n) / norm (f (Suc n)))) \<longlonglongrightarrow> c"
shows "conv_radius f = c"
proof (rule conv_radius_eqI')
show "c \<ge> 0" by (intro Lim_bounded2[OF lim]) simp_all
next
fix r assume r: "0 < r" "ereal r < c"
let ?l = "liminf (\<lambda>n. ereal (norm (f n * of_real r ^ n) / norm (f (Suc n) * of_real r ^ Suc n)))"
have "?l = liminf (\<lambda>n. ereal (norm (f n) / (norm (f (Suc n)))) * ereal (inverse r))"
using r by (simp add: norm_mult norm_power field_split_simps)
also from r have "\<dots> = liminf (\<lambda>n. ereal (norm (f n) / (norm (f (Suc n))))) * ereal (inverse r)"
by (intro Liminf_ereal_mult_right) simp_all
also have "liminf (\<lambda>n. ereal (norm (f n) / (norm (f (Suc n))))) = c"
by (intro lim_imp_Liminf lim) simp
finally have l: "?l = c * ereal (inverse r)" by simp
from r have l': "c * ereal (inverse r) > 1" by (cases c) (simp_all add: field_simps)
show "summable (\<lambda>n. f n * of_real r^n)"
by (rule summable_norm_cancel, rule ratio_test_convergence)
(insert r nz l l', auto elim!: eventually_mono)
next
fix r assume r: "0 < r" "ereal r > c"
let ?l = "limsup (\<lambda>n. ereal (norm (f n * of_real r ^ n) / norm (f (Suc n) * of_real r ^ Suc n)))"
have "?l = limsup (\<lambda>n. ereal (norm (f n) / (norm (f (Suc n)))) * ereal (inverse r))"
using r by (simp add: norm_mult norm_power field_split_simps)
also from r have "\<dots> = limsup (\<lambda>n. ereal (norm (f n) / (norm (f (Suc n))))) * ereal (inverse r)"
by (intro Limsup_ereal_mult_right) simp_all
also have "limsup (\<lambda>n. ereal (norm (f n) / (norm (f (Suc n))))) = c"
by (intro lim_imp_Limsup lim) simp
finally have l: "?l = c * ereal (inverse r)" by simp
from r have l': "c * ereal (inverse r) < 1" by (cases c) (simp_all add: field_simps)
show "\<not>summable (\<lambda>n. norm (f n * of_real r^n))"
by (rule ratio_test_divergence) (insert r nz l l', auto elim!: eventually_mono)
qed
lemma conv_radius_ratio_limit_ereal_nonzero:
fixes f :: "nat \<Rightarrow> 'a :: {banach,real_normed_div_algebra}"
assumes nz: "c \<noteq> 0"
assumes lim: "(\<lambda>n. ereal (norm (f n) / norm (f (Suc n)))) \<longlonglongrightarrow> c"
shows "conv_radius f = c"
proof (rule conv_radius_ratio_limit_ereal[OF _ lim], rule ccontr)
assume "\<not>eventually (\<lambda>n. f n \<noteq> 0) sequentially"
hence "frequently (\<lambda>n. f n = 0) sequentially" by (simp add: frequently_def)
hence "frequently (\<lambda>n. ereal (norm (f n) / norm (f (Suc n))) = 0) sequentially"
by (force elim!: frequently_elim1)
hence "c = 0" by (intro limit_frequently_eq[OF _ _ lim]) auto
with nz show False by contradiction
qed
lemma conv_radius_ratio_limit:
fixes f :: "nat \<Rightarrow> 'a :: {banach,real_normed_div_algebra}"
assumes "c' = ereal c"
assumes nz: "eventually (\<lambda>n. f n \<noteq> 0) sequentially"
assumes lim: "(\<lambda>n. norm (f n) / norm (f (Suc n))) \<longlonglongrightarrow> c"
shows "conv_radius f = c'"
using assms by (intro conv_radius_ratio_limit_ereal) simp_all
lemma conv_radius_ratio_limit_nonzero:
fixes f :: "nat \<Rightarrow> 'a :: {banach,real_normed_div_algebra}"
assumes "c' = ereal c"
assumes nz: "c \<noteq> 0"
assumes lim: "(\<lambda>n. norm (f n) / norm (f (Suc n))) \<longlonglongrightarrow> c"
shows "conv_radius f = c'"
using assms by (intro conv_radius_ratio_limit_ereal_nonzero) simp_all
lemma conv_radius_cmult_left:
assumes "c \<noteq> (0 :: 'a :: {banach, real_normed_div_algebra})"
shows "conv_radius (\<lambda>n. c * f n) = conv_radius f"
proof -
have "conv_radius (\<lambda>n. c * f n) =
inverse (limsup (\<lambda>n. ereal (root n (norm (c * f n)))))"
unfolding conv_radius_def ..
also have "(\<lambda>n. ereal (root n (norm (c * f n)))) =
(\<lambda>n. ereal (root n (norm c)) * ereal (root n (norm (f n))))"
by (rule ext) (auto simp: norm_mult real_root_mult)
also have "limsup \<dots> = ereal 1 * limsup (\<lambda>n. ereal (root n (norm (f n))))"
using assms by (intro ereal_limsup_lim_mult tendsto_ereal LIMSEQ_root_const) auto
also have "inverse \<dots> = conv_radius f" by (simp add: conv_radius_def)
finally show ?thesis .
qed
lemma conv_radius_cmult_right:
assumes "c \<noteq> (0 :: 'a :: {banach, real_normed_div_algebra})"
shows "conv_radius (\<lambda>n. f n * c) = conv_radius f"
proof -
have "conv_radius (\<lambda>n. f n * c) = conv_radius (\<lambda>n. c * f n)"
by (simp add: conv_radius_def norm_mult mult.commute)
with conv_radius_cmult_left[OF assms, of f] show ?thesis by simp
qed
lemma conv_radius_mult_power:
assumes "c \<noteq> (0 :: 'a :: {real_normed_div_algebra,banach})"
shows "conv_radius (\<lambda>n. c ^ n * f n) = conv_radius f / ereal (norm c)"
proof -
have "limsup (\<lambda>n. ereal (root n (norm (c ^ n * f n)))) =
limsup (\<lambda>n. ereal (norm c) * ereal (root n (norm (f n))))"
by (intro Limsup_eq eventually_mono [OF eventually_gt_at_top[of "0::nat"]])
(auto simp: norm_mult norm_power real_root_mult real_root_power)
also have "\<dots> = ereal (norm c) * limsup (\<lambda>n. ereal (root n (norm (f n))))"
using assms by (subst Limsup_ereal_mult_left[symmetric]) simp_all
finally have A: "limsup (\<lambda>n. ereal (root n (norm (c ^ n * f n)))) =
ereal (norm c) * limsup (\<lambda>n. ereal (root n (norm (f n))))" .
show ?thesis using assms
apply (cases "limsup (\<lambda>n. ereal (root n (norm (f n)))) = 0")
apply (simp add: A conv_radius_def)
apply (unfold conv_radius_def A divide_ereal_def, simp add: mult.commute ereal_inverse_mult)
done
qed
lemma conv_radius_mult_power_right:
assumes "c \<noteq> (0 :: 'a :: {real_normed_div_algebra,banach})"
shows "conv_radius (\<lambda>n. f n * c ^ n) = conv_radius f / ereal (norm c)"
using conv_radius_mult_power[OF assms, of f]
unfolding conv_radius_def by (simp add: mult.commute norm_mult)
lemma conv_radius_divide_power:
assumes "c \<noteq> (0 :: 'a :: {real_normed_div_algebra,banach})"
shows "conv_radius (\<lambda>n. f n / c^n) = conv_radius f * ereal (norm c)"
proof -
from assms have "inverse c \<noteq> 0" by simp
from conv_radius_mult_power_right[OF this, of f] show ?thesis
by (simp add: divide_inverse divide_ereal_def assms norm_inverse power_inverse)
qed
lemma conv_radius_add_ge:
"min (conv_radius f) (conv_radius g) \<le>
conv_radius (\<lambda>x. f x + g x :: 'a :: {banach,real_normed_div_algebra})"
by (rule conv_radius_geI_ex')
(auto simp: algebra_simps intro!: summable_add summable_in_conv_radius)
lemma conv_radius_mult_ge:
fixes f g :: "nat \<Rightarrow> ('a :: {banach,real_normed_div_algebra})"
shows "conv_radius (\<lambda>x. \<Sum>i\<le>x. f i * g (x - i)) \<ge> min (conv_radius f) (conv_radius g)"
proof (rule conv_radius_geI_ex')
fix r assume r: "r > 0" "ereal r < min (conv_radius f) (conv_radius g)"
from r have "summable (\<lambda>n. (\<Sum>i\<le>n. (f i * of_real r^i) * (g (n - i) * of_real r^(n - i))))"
by (intro summable_Cauchy_product abs_summable_in_conv_radius) simp_all
thus "summable (\<lambda>n. (\<Sum>i\<le>n. f i * g (n - i)) * of_real r ^ n)"
by (simp add: algebra_simps of_real_def power_add [symmetric] scaleR_sum_right)
qed
lemma le_conv_radius_iff:
fixes a :: "nat \<Rightarrow> 'a::{real_normed_div_algebra,banach}"
shows "r \<le> conv_radius a \<longleftrightarrow> (\<forall>x. norm (x-\<xi>) < r \<longrightarrow> summable (\<lambda>i. a i * (x - \<xi>) ^ i))"
apply (intro iffI allI impI summable_in_conv_radius conv_radius_geI_ex)
apply (meson less_ereal.simps(1) not_le order_trans)
apply (rule_tac x="of_real ra" in exI, simp)
apply (metis abs_of_nonneg add_diff_cancel_left' less_eq_real_def norm_of_real)
done
end