Theory Lambert_Series
section ‹Lambert Series›
theory Lambert_Series
imports
"HOL-Complex_Analysis.Complex_Analysis"
"HOL-Real_Asymp.Real_Asymp"
"Dirichlet_Series.Dirichlet_Series_Analysis"
"Dirichlet_Series.Divisor_Count"
Polylog.Polylog
Lambert_Series_Library
Number_Theoretic_Functions_Extras
Summation_Tests_More
begin
subsection ‹Definition›
no_notation Infinite_Set_Sum.abs_summable_on (infix ‹abs'_summable'_on› 50)
text ‹
Given any sequence $a(n)$ for $n \geq 1$, the corresponding ∗‹Lambert series› is defined as
\[L(a, q) = \sum_{n=1}^\infty a(n) \frac{q^n}{1-q^n}\ .\]
›
definition lambert :: "(nat ⇒ 'a :: {real_normed_field, banach}) ⇒ 'a ⇒ 'a" where
"lambert a q =
(let f = (λn. a (Suc n) * q ^ (Suc n) / (1 - q ^ (Suc n))) in
if summable f then ∑n. f n else 0)"
lemma lambert_eqI:
assumes "(λn. a (Suc n) * q ^ (Suc n) / (1 - q ^ (Suc n))) sums x"
shows "lambert a q = x"
using assms unfolding lambert_def Let_def sums_iff by simp
lemma lambert_cong [cong]:
"(⋀n. n > 0 ⟹ a n = a' n) ⟹ q = q' ⟹ lambert a q = lambert a' q'"
by (simp add: lambert_def)
lemma lambert_0 [simp]: "lambert a 0 = 0"
by (simp add: lambert_def)
lemma lambert_0' [simp]: "lambert (λ_. 0) q = 0"
by (simp add: lambert_def)
lemma lambert_cmult: "lambert (λn. c * a n) q = c * lambert a q"
proof (cases "c = 0")
case False
define f where "f = (λn. a (Suc n) * q ^ (Suc n) / (1 - q ^ (Suc n)))"
show ?thesis
proof (cases "summable f")
case True
hence "(λn. c * (a (Suc n) * q ^ (Suc n) / (1 - q ^ (Suc n)))) sums (c * (∑n. f n))"
unfolding mult.assoc by (intro sums_mult) (auto simp: f_def)
thus ?thesis using True
by (intro lambert_eqI) (auto simp: lambert_def f_def algebra_simps)
next
case False
hence "¬summable (λn. c * f n)"
using ‹c ≠ 0› by simp
with False show ?thesis
by (simp add: lambert_def f_def algebra_simps)
qed
qed auto
lemma lambert_cmult': "lambert (λn. a n * c) q = lambert a q * c"
using lambert_cmult[of c a q] by (simp add: mult_ac)
lemma lambert_uminus: "lambert (λn. -a n) q = -lambert a q"
using lambert_cmult[of "-1" a q] by simp
text ‹
We will later see that if $\sum_{n=1}^\infty a(n)$ exists then the Lambert series converges
everywhere except on the unit circle; otherwise it has the same convergence radius as $a$
(and that radius then has to be ${<}\,1$).
›
definition lambert_conv_radius :: "(nat ⇒ 'a :: {banach, real_normed_field}) ⇒ ereal"
where "lambert_conv_radius a = (if summable a then ∞ else conv_radius a)"
lemma lambert_conv_radius_gt_1_iff: "lambert_conv_radius a > 1 ⟷ summable a"
proof
assume *: "lambert_conv_radius a > 1"
{
assume "¬summable a"
hence "conv_radius a > 1"
using * by (auto simp: lambert_conv_radius_def)
hence "summable (λn. a n * 1 ^ n)"
by (intro summable_in_conv_radius) (auto simp: one_ereal_def)
with ‹¬summable a› have False
by simp
}
thus "summable a"
by blast
qed (auto simp: lambert_conv_radius_def)
subsection ‹Uniform convergence, continuity, holomorphicity›
text ‹
We will now show some (uniform) convergence results for $L(a, q)$, which will then give us
the holomorphicity and continuity of $L(a, q)$. We will also show some absolute summability
results.
›
context
fixes a :: "nat ⇒ 'a :: {real_normed_field, banach}"
fixes f :: "nat ⇒ 'a ⇒ 'a" and A :: "'a"
defines "f ≡ λk q. a k * q ^ k / (1 - q ^ k)"
defines "A ≡ (∑n. a (Suc n))"
begin
text ‹
Let $a(n)$ have convergence radius $r$. In discs of radius $\text{min}(1, r)$, the Lambert
series for $a(n)$ converges uniformly. This is a simple application of Weierstraß's $M$ test.
›
lemma uniform_limit_lambert1_aux:
fixes r :: real
assumes "0 < r" "r < min 1 (conv_radius a)"
shows "uniform_limit (ball 0 r) (λn q. (∑k<n. f (Suc k) q)) (λq. ∑k. f (Suc k) q) sequentially"
proof -
from assms have r: "r > 0" "r < 1" "r < conv_radius a"
by auto
show "uniform_limit (ball 0 r) (λn q. (∑k<n. f (Suc k) q)) (λq. ∑k. f (Suc k) q) sequentially"
proof (rule Weierstrass_m_test_ev)
have "eventually (λk. 1 - r ^ k ≥ 1 / 2) at_top"
using r by real_asymp
hence "eventually (λk. ∀q∈ball 0 r. norm (f k q) ≤ 2 * norm (a k) * r ^ k) at_top"
using eventually_gt_at_top[of 0]
proof eventually_elim
case k: (elim k)
show "∀q∈ball 0 r. norm (f k q) ≤ 2 * norm (a k) * r ^ k"
proof
fix q :: 'a assume q: "q ∈ ball 0 r"
have "norm (f k q) = norm (a k) * norm q ^ k / norm (1 - q ^ k)"
by (simp add: f_def norm_mult norm_divide norm_power)
also {
have "1 / 2 ≤ 1 - r ^ k"
using k by simp
also have "… ≤ norm (1 :: 'a) - norm (q ^ k)"
using q by (auto simp: norm_power intro!: power_mono)
also have "… ≤ norm (1 - q ^ k)"
by norm
finally have "norm (1 - q ^ k) ≥ 1 / 2" .
}
hence "norm (a k) * norm q ^ k / norm (1 - q ^ k) ≤
norm (a k) * r ^ k / (1 / 2)"
using q r k
by (intro mult_mono power_mono frac_le)
(auto intro!: mult_pos_pos simp: power_less_1_iff norm_power dest!: power_eq_1_iff)
finally show "norm (f k q) ≤ 2 * norm (a k) * r ^ k"
by simp
qed
qed
thus "∀⇩F k in sequentially. ∀q∈ball 0 r. norm (f (Suc k) q) ≤ 2 * norm (a (Suc k)) * r ^ Suc k"
by (rule eventually_compose_filterlim[OF _ filterlim_Suc])
next
have "summable (λk. 2 * (norm (a (Suc k) * of_real r ^ Suc k)))"
by (subst summable_Suc_iff, intro summable_mult abs_summable_in_conv_radius) (use r in auto)
thus "summable (λk. 2 * norm (a (Suc k)) * r ^ Suc k)"
using ‹r > 0› by (simp add: norm_mult norm_power mult_ac)
qed
qed
lemma uniform_limit_lambert1:
fixes r :: real
assumes "0 < r" "r < min 1 (conv_radius a)"
shows "uniform_limit (ball 0 r) (λn q. (∑k<n. f (Suc k) q)) (lambert a) sequentially"
proof -
have lim: "uniform_limit (ball 0 r) (λn q. (∑k<n. f (Suc k) q)) (λq. ∑k. f (Suc k) q) sequentially"
using assms by (rule uniform_limit_lambert1_aux)
also have "?this ⟷ ?thesis"
proof (intro uniform_limit_cong ballI allI refl always_eventually)
fix q :: 'a assume q: "q ∈ ball 0 r"
have *: "(λn. a (Suc n) * q ^ Suc n / (1 - q ^ Suc n)) sums (∑k. f (Suc k) q)"
using tendsto_uniform_limitI[OF lim q] unfolding f_def by (simp add: sums_def)
show "(∑k. f (Suc k) q) = lambert a q"
by (rule sym, rule lambert_eqI) (fact *)
qed
finally show ?thesis .
qed
text ‹
Since $a_n \frac{q^n}{1-q^n} = -a_n - a_n\frac{(\frac{1}{q})^n}{1-(\frac{1}{q})^n}$,
we can substitute $q \mapsto \frac{1}{q}$ in the above uniform convergence result to deduce
that uniform convergence also holds on any annulus $r \leq |q| \leq R$ with $1 < r < R$.
›
lemma uniform_limit_lambert2:
fixes r R :: real
assumes r: "1 < r" "r < R"
assumes "summable a"
defines "D ≡ cball 0 R - ball 0 r"
shows "uniform_limit D (λn q. (∑k<n. f (Suc k) q)) (λq. -A - lambert a (1 / q)) sequentially"
proof -
define g where "g = (λn q. a n * (1 / q) ^ n / (1 - (1 / q) ^ n))"
from r(1) obtain r' where r': "1 < r'" "r' < r"
using dense by blast
have "uniform_limit D (λn q. ∑k<n. f (Suc k) (1 / q)) (λq. lambert a (1 / q)) sequentially"
using uniform_limit_lambert1
proof (rule uniform_limit_compose[where g = "λq. 1 / q"])
have "conv_radius a ≥ norm (of_real 1 :: 'a)"
by (rule conv_radius_geI) (use ‹summable a› in auto)
hence "min 1 (conv_radius a) = 1"
by (simp add: min_def one_ereal_def)
with r' show "1 / r' > 0" "ereal (1 / r') < min 1 (conv_radius a)"
by auto
next
show "1 / q ∈ ball 0 (1 / r')" if "q ∈ D" for q
using that r r' by (auto simp: D_def norm_divide divide_simps not_less)
qed
moreover have "summable (λn. a (Suc n))"
using ‹summable a› by (simp add: summable_Suc_iff)
hence "(λn. ∑k<n. a (Suc k)) ⇢ A"
unfolding A_def summable_sums_iff sums_def .
ultimately have "uniform_limit D (λn q. -(∑k<n. a (Suc k)) - (∑k<n. f (Suc k) (1/q)))
(λq. -A - lambert a (1 / q)) sequentially"
by (intro uniform_limit_intros uniform_limit_const')
also have "?this ⟷ ?thesis"
proof (intro uniform_limit_cong refl always_eventually allI ballI)
fix n :: nat and q :: 'a
assume q: "q ∈ D"
hence q': "q ≠ 0"
using r by (auto simp: D_def)
have q'': "q ^ k ≠ 1" if "k > 0" for k
using q r that by (auto dest!: power_eq_1_iff simp: D_def)
have "-(∑k<n. a (Suc k)) - (∑k<n. f (Suc k) (1 / q)) =
(∑k<n. -a (Suc k) - f (Suc k) (1 / q))"
by (simp add: sum_negf sum_subtractf)
also have *: "-a k - f k (1 / q) = f k q" if "k > 0" for k
using that q' q''[of k] by (auto simp: f_def field_simps)
have "(∑k<n. -a (Suc k) - f (Suc k) (1 / q)) = (∑k<n. f (Suc k) q)"
by (intro sum.cong refl *) auto
finally show "-(∑k<n. a (Suc k)) - (∑k<n. f (Suc k) (1 / q)) = (∑k<n. f (Suc k) q)" .
qed
finally show ?thesis .
qed
text ‹
With some more book-keeping, we show that the series converges uniformly in all compact
sets that do not touch the unit circle and, if $\sum_{n=1}^\infty a(n)$ does not exist,
lie fully within the convergence radius of $a(n)$. This is mentioned in Knopp's Theorem~259.
›
theorem uniform_limit_lambert:
assumes "compact X" "X ⊆ eball 0 (lambert_conv_radius a) - sphere 0 1"
shows "uniform_limit X (λn q. (∑k<n. f (Suc k) q)) (lambert a) sequentially"
proof -
from assms have norm_neq_1: "norm x ≠ 1" if "x ∈ X" for x
using that by auto
have 1: "uniform_limit (X ∩ cball 0 1) (λn q. (∑k<n. f (Suc k) q)) (lambert a) sequentially"
proof (cases "X ∩ cball 0 1 ⊆ {0}")
case True
hence "X ∩ cball 0 1 = {} ∨ X ∩ cball 0 1 = {0}"
by auto
thus ?thesis
by (elim disjE) (simp_all add: f_def lambert_def)
next
case False
obtain r :: real
where r: "r > 0" "r < 1" "r < conv_radius a" "⋀x. x ∈ X ∩ cball 0 1 ⟹ norm x < r"
proof -
define c where "c = (if lambert_conv_radius a > 1 then 1
else real_of_ereal (lambert_conv_radius a))"
have "compact ((ereal ∘ norm) ` (X ∩ cball 0 1))"
by (intro compact_continuous_image continuous_intros compact_insert) (use assms in auto)
hence "Sup ((ereal ∘ norm) ` (X ∩ cball 0 1)) ∈ (ereal ∘ norm) ` (X ∩ cball 0 1)"
using False by (intro closed_contains_Sup_cl) (auto intro: compact_imp_closed)
then obtain q where q: "q ∈ (X ∩ cball 0 1)"
"ereal (norm q) = Sup ((ereal ∘ norm) ` (X ∩ cball 0 1))"
unfolding o_def by force
have q': "norm q' ≤ norm q" if "q' ∈ X ∩ cball 0 1" for q'
using Sup_upper q that unfolding o_def by (metis ereal_less_eq(3) imageI insertCI)
have "q ≠ 0"
using q' False by force
have "conv_radius a ≥ norm (1 :: 'a)" if "summable a"
by (rule conv_radius_geI) (use that in auto)
hence "norm q < conv_radius a" "norm q < 1"
using q(1) assms ‹q ≠ 0›
by (auto simp: lambert_conv_radius_def eball_def ereal_le_less split: if_splits)
then obtain r where r: "norm q < r" "r < 1" "r < conv_radius a"
by (smt (verit, ccfv_SIG) ereal_dense2 ereal_less(3) less_ereal.simps(1) linorder_not_le order_less_le_trans)
show ?thesis
proof (rule that[of r])
show "r > 0"
using r q ‹q ≠ 0› by (smt (verit) norm_ge_zero)
show "r < 1" "ereal r < conv_radius a"
using r by auto
show "norm q' < r" if "q' ∈ X ∩ cball 0 1" for q'
using assms q q' that ‹q ≠ 0› r
by (force simp: lambert_conv_radius_def eball_def split: if_splits)
qed
qed
have "uniform_limit (ball 0 r) (λn q. (∑k<n. f (Suc k) q)) (lambert a) sequentially"
by (rule uniform_limit_lambert1) (use r in auto)
thus "uniform_limit (X ∩ cball 0 1) (λn q. (∑k<n. f (Suc k) q)) (lambert a) sequentially"
by (rule uniform_limit_on_subset) (use r in auto)
qed
have 2: "uniform_limit (X - ball 0 1) (λn q. (∑k<n. f (Suc k) q)) (lambert a) sequentially"
proof (cases "∃q∈X. norm q > 1")
case False
hence *: "X - ball 0 1 = {}"
using norm_neq_1 by fastforce
show ?thesis
unfolding * by simp
next
case True
then obtain q where q: "q ∈ X" "norm q > 1"
by auto
with assms have "lambert_conv_radius a > 1"
by (smt (verit, ccfv_SIG) Diff_subset dist_0_norm ereal_less(3) in_eball_iff linorder_not_le order_less_le_trans subsetD)
hence "summable a"
by (simp add: lambert_conv_radius_gt_1_iff)
obtain r where r: "r > 1" "⋀x. x ∈ X - cball 0 1 ⟹ norm x > r"
proof -
have compact: "compact (norm ` (X - ball 0 1))"
by (intro compact_continuous_image compact_diff continuous_intros) (use assms in auto)
hence "Inf (norm ` (X - ball 0 1)) ∈ norm ` (X - ball 0 1)"
using q by (intro closed_contains_Inf)
(auto intro: compact_imp_closed bounded_imp_bdd_below compact_imp_bounded)
then obtain q' where q': "q' ∈ X - ball 0 1" "norm q' = Inf (norm ` (X - ball 0 1))"
by force
have "norm q' > 1"
using q' assms by auto
then obtain r where r: "1 < r" "r < norm q'"
using dense by blast
show ?thesis
proof (rule that[of r])
show "r > 1"
using q' assms r by auto
show "norm q'' > r" if "q'' ∈ X - cball 0 1" for q''
proof -
have "r < Inf (norm ` (X - ball 0 1))"
using r q' by simp
also have "… ≤ norm q''"
by (rule cInf_lower)
(use that assms compact in ‹auto intro!: bounded_imp_bdd_below compact_imp_bounded›)
finally show ?thesis .
qed
qed
qed
obtain R where R: "R > r" "⋀x. x ∈ X ⟹ norm x < R"
proof -
have "bounded X"
using assms compact_imp_bounded by blast
then obtain R where R: "norm x ≤ R" if "x ∈ X" for x
unfolding bounded_iff by blast
show ?thesis
proof (rule that[of "R + 1"])
show "r < R + 1"
using r(2)[of q] q R[of q] by auto
show "norm x < R + 1" if "x ∈ X" for x
using R[of x] that by auto
qed
qed
have lim: "uniform_limit (cball 0 R - ball 0 r) (λn q. (∑k<n. f (Suc k) q))
(λq. -A - lambert a (1 / q)) sequentially"
by (rule uniform_limit_lambert2) (use r R ‹summable a› in auto)
also have "?this ⟷ uniform_limit (cball 0 R - ball 0 r) (λn q. (∑k<n. f (Suc k) q))
(lambert a) sequentially"
proof (intro uniform_limit_cong refl always_eventually allI ballI)
fix q :: 'a assume q: "q ∈ cball 0 R - ball 0 r"
with lim have "(λn. (∑k<n. f (Suc k) q)) ⇢ -A - lambert a (1 / q)"
by (rule tendsto_uniform_limitI)
hence "(λk. f (Suc k) q) sums (-A - lambert a (1 / q))"
by (simp add: sums_def)
thus "-A - lambert a (1 / q) = lambert a q"
unfolding lambert_def[of a q] by (simp add: sums_iff f_def)
qed
finally show ?thesis
by (rule uniform_limit_on_subset) (use r R norm_neq_1 in force)+
qed
have "uniform_limit ((X ∩ cball 0 1) ∪ (X - ball 0 1))
(λn q. (∑k<n. f (Suc k) q)) (lambert a) sequentially"
using 1 2 by (rule uniform_limit_on_Un)
also have "(X ∩ cball 0 1) ∪ (X - ball 0 1) = X"
using norm_neq_1 by auto
finally show ?thesis .
qed
lemma sums_lambert:
assumes "norm q < lambert_conv_radius a" "norm q ≠ 1"
shows "(λk. f (Suc k) q) sums lambert a q"
proof -
have "(λn. (∑k<n. f (Suc k) q)) ⇢ lambert a q"
proof (rule tendsto_uniform_limitI[OF uniform_limit_lambert])
show "compact {q}" "{q} ⊆ eball 0 (lambert_conv_radius a) - sphere 0 1"
using assms by auto
qed auto
thus ?thesis
by (simp add: sums_def)
qed
text ‹
A side effect of this: the functional equation
\[L(a, \tfrac{1}{q}) = -\big(\sum\nolimits_{n=1}^\infty a(n)\big) - L(a, q)\ ,\]
which is valid for all $q$ with $q \neq 0$ and $|q| \neq 1$ if $\sum_{n=1}^\infty a(n)$ exists.
›
theorem lambert_reciprocal:
assumes "summable a" and "q ≠ 0" and "norm q ≠ 1"
shows "lambert a (1 / q) = -A - lambert a q"
proof -
have *: "lambert a (1 / q) = -A - lambert a q" if q: "norm q > 1" for q
proof -
obtain r where r: "1 < r" "r < norm q"
using q dense by blast
have "uniform_limit (cball 0 (norm q + 1) - ball 0 r)
(λn q. ∑k<n. f (Suc k) q)
(λq. - A - lambert a (1 / q)) sequentially"
by (rule uniform_limit_lambert2) (use assms r in auto)
hence "(λk. f (Suc k) q) sums (-A - lambert a (1 / q))"
unfolding sums_def by (rule tendsto_uniform_limitI) (use r in auto)
moreover have "uniform_limit {q} (λn q. ∑k<n. f (Suc k) q) (lambert a) sequentially"
using assms r
by (intro uniform_limit_lambert compact_diff compact_cball)
(auto simp: lambert_conv_radius_def)
hence "(λk. f (Suc k) q) sums lambert a q"
unfolding sums_def by (rule tendsto_uniform_limitI) (use r in auto)
ultimately show ?thesis
by (simp add: sums_iff algebra_simps)
qed
show ?thesis
proof (cases "norm q > 1")
case False
thus ?thesis using assms
using *[of "1 / q"] by (auto simp: norm_divide)
qed (use *[of q] in auto)
qed
lemma summable_lambert:
assumes "norm q < lambert_conv_radius a" "norm q ≠ 1"
shows "summable (λk. f k q)"
using sums_lambert[OF assms] unfolding sums_iff by (subst (asm) summable_Suc_iff) auto
text ‹
We have shown that the Lambert series for $a(n)$ converges everywhere except on the unit circle if
$\sum_{n=1}^\infty a(n)$ exists, and it converges within the convergence radius of $R$ of $a(n)$
otherwise.
We will now show that within $\text{min}(1, R)$, this convergence is absolute.
›
lemma norm_summable_lambert:
assumes "norm q < min 1 (conv_radius a)"
shows "summable (λk. norm (f k q))"
proof (cases "q = 0")
case [simp]: True
have "eventually (λk. norm (f k q) = 0) at_top"
using eventually_gt_at_top[of 0] by eventually_elim (auto simp: f_def)
thus ?thesis
using summable_cong by fastforce
next
case False
define R where "R = (if conv_radius a > 1 then 1 else real_of_ereal (conv_radius a))"
have R: "ereal R = min 1 (conv_radius a)"
using conv_radius_nonneg[of a] by (cases "conv_radius a") (auto simp: R_def)
with assms have "norm q < R"
by (metis less_ereal.simps(1))
then obtain r where r: "norm q < r" "r < R"
using dense by blast
hence "r > 0"
using norm_ge_zero le_less_trans by blast
have "r < 1"
using r R by (metis ereal_less(3) less_ereal.simps(1) min_less_iff_conj)
have "r < conv_radius a"
using r R by (metis less_ereal.simps(1) min_less_iff_conj)
have "norm q < 1"
using assms by auto
note r' = ‹r > 0› ‹r < 1› ‹r < conv_radius a›
show ?thesis
proof (rule summable_powser_comparison_test_bigo)
show "summable (λn. a n * of_real r ^ n)"
by (rule summable_in_conv_radius) (use r r' R in auto)
next
have "(λn. norm (f n q)) = (λn. norm (a n) * norm q ^ n / norm (1 - q ^ n))"
by (simp add: f_def norm_mult norm_divide norm_power)
also have "1 - norm q ^ n ≤ norm (1 - q ^ n)" for n
by (metis norm_one norm_power norm_triangle_ineq2)
hence "(λn. norm (a n) * norm q ^ n / norm (1 - q ^ n)) ∈
O(λn. norm (a n) * (norm q ^ n / (1 - norm q ^ n)))"
using ‹q ≠ 0› ‹norm q < 1›
by (intro bigoI[of _ 1] eventually_mono[OF eventually_gt_at_top[of 0]])
(auto intro!: mult_left_mono divide_left_mono mult_pos_pos add_pos_pos
dest!: power_eq_1_iff simp: power_le_one power_less_1_iff)
also have "(λn. norm (a n) * (norm q ^ n / (1 - norm q ^ n))) ∈ O(λn. norm (a n) * norm q ^ n)"
by (intro landau_o.big.mult_left) (use ‹q ≠ 0› ‹norm q < 1› in real_asymp)
also have "(λn. norm (a n) * norm q ^ n) =
(λn. norm (a n * of_real r ^ n * (q / of_real r) ^ n))"
using ‹r > 0› by (intro ext) (auto simp: norm_mult norm_divide norm_power power_divide)
finally show "(λn. f n q) ∈ O(λn. a n * of_real r ^ n * (q / of_real r) ^ n)"
by (subst (asm) landau_o.big.norm_iff)
next
show "norm (q / of_real r) < 1"
using r r' by (simp add: norm_divide field_simps)
qed
qed
text ‹
If additionally $\sum_{k=1}^\infty a(k)$ converges absolutely, the absolute convergence of the
Lambert series also holds everywhere.
›
lemma norm_summable_lambert':
assumes "summable (λk. norm (a k))" and "norm q ≠ 1"
shows "summable (λk. norm (f k q))"
proof -
have *: "summable (λk. norm (f k q))" if q: "norm q < 1" for q
proof -
have "conv_radius a ≥ norm (1 :: 'a)"
using assms(1) by (intro conv_radius_geI) (auto dest: summable_norm_cancel)
with q have "ereal (norm q) < conv_radius a"
by (simp add: ereal_le_less)
thus ?thesis
using assms(2) q by (intro norm_summable_lambert) auto
qed
show ?thesis
proof (cases "norm q < 1")
case True
thus ?thesis using *[of q] by simp
next
case False
hence [simp]: "q ≠ 0"
by auto
have [simp]: "q ^ k = 1 ⟷ k = 0" for k
using assms(2) by (auto dest: power_eq_1_iff)
have "summable (λk. norm (a k + f k (inverse q)))"
using False assms(2) by (intro summable_norm_add assms *) (auto simp: norm_divide field_simps)
moreover have "eventually (λk. f k q = -a k - f k (inverse q)) at_top"
using eventually_gt_at_top[of 0]
by eventually_elim (use False assms(2) in ‹auto simp: fun_eq_iff field_simps f_def›)
hence "eventually (λk. norm (f k q) = norm (a k + f k (inverse q))) at_top"
by eventually_elim (auto simp: norm_uminus_minus)
ultimately show ?thesis
using summable_cong by fast
qed
qed
lemma abs_summable_on_lambert:
assumes "norm q < min 1 (conv_radius a)"
shows "(λk. f k q) abs_summable_on {1..}"
proof -
have "summable (λk. norm (f (Suc k) q))"
by (subst summable_Suc_iff, rule norm_summable_lambert) (use assms in auto)
hence "(λk. f (Suc k) q) abs_summable_on UNIV"
by (subst summable_on_UNIV_nonneg_real_iff) auto
also have "?this ⟷ ?thesis"
by (intro summable_on_reindex_bij_witness[of _ "λk. k - 1" Suc]) auto
finally show ?thesis .
qed
lemma abs_summable_on_lambert':
assumes "summable (λk. norm (a k))" and "norm q ≠ 1"
shows "(λk. f k q) abs_summable_on {1..}"
proof -
have "summable (λk. norm (f (Suc k) q))"
by (subst summable_Suc_iff, rule norm_summable_lambert') (use assms in auto)
hence "(λk. f (Suc k) q) abs_summable_on UNIV"
by (subst summable_on_UNIV_nonneg_real_iff) auto
also have "?this ⟷ ?thesis"
by (intro summable_on_reindex_bij_witness[of _ "λk. k - 1" Suc]) auto
finally show ?thesis .
qed
lemma summable_on_lambert:
assumes "norm q < min 1 (conv_radius a)"
shows "(λk. f k q) summable_on {1..}"
using abs_summable_summable[OF abs_summable_on_lambert[OF assms]] .
lemma has_sum_lambert:
assumes "norm q < min 1 (conv_radius a)"
shows "((λk. f k q) has_sum lambert a q) {1..}"
proof -
have "((λk. f (Suc k) q) has_sum lambert a q) UNIV"
proof (rule norm_summable_imp_has_sum)
show "summable (λn. norm (f (Suc n) q))"
using norm_summable_lambert[OF assms] by (subst summable_Suc_iff)
show "(λk. f (Suc k) q) sums lambert a q"
by (rule sums_lambert) (use assms in ‹auto simp: lambert_conv_radius_def›)
qed
also have "?this ⟷ ?thesis"
by (intro has_sum_reindex_bij_witness[of _ "λk. k - 1" Suc]) auto
finally show ?thesis .
qed
text ‹
We can also show a more precise convergence result that essentially fully reduces the question
of convegence of a Lambert series to that of its ``corresponding'' power series:
$\sum_{k=1}^\infty a(k) \frac{q^k}{1-q^k}$ converges if and only if the ``corresponding''
power series $\sum_{k=1}^\infty a(k) q^k$ converges or if $\sum_{k=1}^\infty a(k)$ converges.
This is Theorem~259 in Knopp's book. A key ingredient, aside from the results we have amassed
so far, is the du-Bois Reymond summation test.
›
theorem summable_lambert_iff:
assumes "norm q ≠ 1"
shows "summable (λk. f k q) ⟷ summable a ∨ summable (λk. a k * q ^ k)"
proof (cases "summable a")
case True
hence "summable (λk. f k q)" using assms
by (intro summable_lambert) (auto simp: lambert_conv_radius_def)
with True show ?thesis
by auto
next
case not_summable: False
have [simp]: "q ^ k ≠ 1" if "k > 0" for k
using that assms by (auto dest: power_eq_1_iff)
have "summable (λk. f k q) ⟷ summable (λk. a k * q ^ k)"
proof (cases "norm q < 1")
case False
with assms have q: "norm q > 1"
by simp
have "¬summable (λk. a k * q ^ k)"
using q not_summable conv_radius_geI[of a q] summable_in_conv_radius[of 1 a]
by (auto simp: ereal_le_less one_ereal_def)
moreover have "¬summable (λk. f k q)"
proof
assume "summable (λk. f k q)"
hence *: "summable (λk. a k / (1 - q ^ k) * q ^ k)"
by (auto simp: f_def)
hence "summable (λk. a k / (1 - q ^ k) * 1 ^ k)"
by (rule powser_inside) (use q in auto)
hence "summable (λk. a k / (1 - q ^ k) - a k / (1 - q ^ k) * q ^ k)"
by (intro summable_diff *) auto
moreover have "eventually (λk. a k / (1 - q ^ k) - a k / (1 - q ^ k) * q ^ k = a k) at_top"
using eventually_gt_at_top[of 0]
by eventually_elim (simp add: divide_simps, simp add: algebra_simps)
ultimately have "summable a"
using summable_cong by fast
with not_summable show False
by contradiction
qed
ultimately show ?thesis
by simp
next
case q: True
show ?thesis
proof
assume "summable (λk. f k q)"
hence "summable (λk. f k q * (1 - q ^ k))"
proof (rule dubois_reymond_summation_test)
have "summable (λk. norm (q - 1) * norm q ^ k)"
using q by (intro summable_mult summable_geometric) auto
also have "(λk. norm (q - 1) * norm q ^ k) = (λk. norm ((q - 1) * q ^ k))"
by (simp add: norm_mult norm_power)
also have "… = (λk. norm (1 - q ^ k - (1 - q ^ Suc k)))"
by (simp add: algebra_simps)
finally show "summable (λk. norm (1 - q ^ k - (1 - q ^ Suc k)))" .
qed
moreover have "eventually (λk. f k q * (1 - q ^ k) = a k * q ^ k) at_top"
using eventually_gt_at_top[of 0] by eventually_elim (auto simp: divide_simps f_def)
ultimately show "summable (λk. a k * q ^ k)"
using summable_cong by fast
next
assume "summable (λk. a k * q ^ k)"
hence "summable (λk. a k * q ^ k * (1 / (1 - q ^ k)))"
proof (rule dubois_reymond_summation_test)
show "summable (λk. norm (1 / (1 - q ^ k) - 1 / (1 - q ^ Suc k)))"
proof (rule summable_comparison_test_ev)
show "summable (λk. 2 / (1 - norm q) ^ 2 * norm q ^ k)"
using q by (intro summable_mult summable_geometric) auto
next
show "eventually (λk. norm (norm (1 / (1 - q ^ k) - 1 / (1 - q ^ Suc k))) ≤
2 / (1 - norm q) ^ 2 * norm q ^ k) at_top"
using eventually_gt_at_top[of 0]
proof eventually_elim
case k: (elim k)
have "norm (1 - q) ≤ norm (1 :: 'a) + 1"
using q by norm
hence 1: "norm (1 - q) ≤ 2"
by simp
have 2: "norm (1 - q ^ l) ≥ 1 - norm q" if l: "l > 0" for l
proof -
have "norm (1 - q ^ l) ≥ norm (1 :: 'a) - norm (q ^ l)"
by norm
moreover have "norm (q ^ l) ≤ norm q ^ 1"
using q l unfolding norm_power by (intro power_decreasing) auto
ultimately show ?thesis
by simp
qed
have "1 / (1 - q ^ k) - 1 / (1 - q ^ Suc k) =
(q ^ k - q ^ Suc k) / ((1 - q ^ k) * (1 - q ^ Suc k))"
using k by (simp add: divide_simps del: power_Suc)
also have "… = (1 - q) * q ^ k / ((1 - q ^ k) * (1 - q ^ Suc k))"
by (simp add: algebra_simps)
also have "norm … = norm (1 - q) * norm q ^ k / (norm (1 - q ^ k) * norm (1 - q ^ Suc k))"
by (simp add: norm_mult norm_divide norm_power)
also have "… ≤ 2 * norm q ^ k / ((1 - norm q) * (1 - norm q))" using q k
by (intro frac_le mult_mono mult_pos_pos zero_le_power norm_ge_zero 1 2)
(auto simp del: power_Suc)
finally show ?case
by (simp add: power2_eq_square)
qed
qed
qed
thus "summable (λk. f k q)"
by (simp add: f_def)
qed
qed
with not_summable show ?thesis
by blast
qed
end
lemma holomorphic_lambert [holomorphic_intros]:
assumes "X ⊆ eball 0 (lambert_conv_radius a) - sphere 0 1"
shows "lambert a holomorphic_on X"
proof -
have "lambert a holomorphic_on eball 0 (lambert_conv_radius a) - sphere 0 1"
proof (rule holomorphic_uniform_sequence)
show "open (eball 0 (lambert_conv_radius a) - sphere (0 :: complex) 1)"
by (intro open_Diff) auto
next
fix q :: complex
assume q: "q ∈ eball 0 (lambert_conv_radius a) - sphere 0 1"
have "open (eball 0 (lambert_conv_radius a) - sphere 0 1 :: complex set)"
by auto
then obtain r where r: "r > 0" "cball q r ⊆ eball 0 (lambert_conv_radius a) - sphere 0 1"
unfolding open_contains_cball using q by blast
define f where "f ≡ λk q. a k * q ^ k / (1 - q ^ k)"
show "∃d>0. cball q d ⊆ eball 0 (lambert_conv_radius a) - sphere 0 1 ∧
uniform_limit (cball q d) (λn q. ∑k<n. f (Suc k) q) (lambert a) sequentially"
proof (intro exI[of _ r] conjI)
show "uniform_limit (cball q r) (λn q. ∑k<n. f (Suc k) q) (lambert a) sequentially"
unfolding f_def by (rule uniform_limit_lambert) (use r in auto)
qed (use r in auto)
next
show "(λq. ∑k<n. a (Suc k) * q ^ Suc k / (1 - q ^ Suc k)) holomorphic_on
eball 0 (lambert_conv_radius a) - sphere 0 1" for n :: nat
by (intro holomorphic_intros) (auto simp del: power_Suc dest!: power_eq_1_iff)
qed
thus ?thesis
by (rule holomorphic_on_subset) fact
qed
lemma holomorphic_lambert' [holomorphic_intros]:
assumes "f holomorphic_on A" "⋀z. z ∈ A ⟹ f z ∈ eball 0 (lambert_conv_radius a) - sphere 0 1"
shows "(λz. lambert a (f z)) holomorphic_on A"
by (rule holomorphic_on_compose_gen[OF assms(1) holomorphic_lambert[OF order.refl],
unfolded o_def])
(use assms(2) in auto)
lemma analytic_lambert [analytic_intros]:
fixes a :: "nat ⇒ complex"
assumes "A ⊆ eball 0 (lambert_conv_radius a) - sphere 0 1"
shows "lambert a analytic_on A"
proof -
have "open (eball 0 (lambert_conv_radius a) - sphere 0 1 :: complex set)"
by auto
hence "lambert a analytic_on eball 0 (lambert_conv_radius a) - sphere 0 1"
using holomorphic_lambert[OF order.refl, of a]
by (auto simp: analytic_on_open)
thus ?thesis
by (rule analytic_on_subset) fact
qed
lemma analytic_lambert' [analytic_intros]:
assumes "f analytic_on A" "⋀z. z ∈ A ⟹ f z ∈ eball 0 (lambert_conv_radius a) - sphere 0 1"
shows "(λz. lambert a (f z)) analytic_on A"
by (rule analytic_on_compose_gen[OF assms(1) analytic_lambert[OF order.refl], unfolded o_def])
(use assms(2) in auto)
lemma continuous_on_lambert [continuous_intros]:
fixes a :: "nat ⇒ 'a :: {real_normed_field, banach, heine_borel}"
assumes "A ⊆ eball 0 (lambert_conv_radius a) - sphere 0 1"
shows "continuous_on A (lambert a)"
proof -
have "isCont (lambert a) q" if q: "q ∈ eball 0 (lambert_conv_radius a) - sphere 0 1" for q
proof -
have "open (eball 0 (lambert_conv_radius a) - sphere 0 1 :: 'a set)"
by auto
with q obtain r where r: "r > 0" "cball q r ⊆ eball 0 (lambert_conv_radius a) - sphere 0 1"
unfolding open_contains_cball by blast
have "continuous_on (cball q r) (lambert a)"
proof (rule uniform_limit_theorem)
show "uniform_limit (cball q r)
(λn q. ∑k<n. a (Suc k) * q ^ Suc k / (1 - q ^ Suc k)) (lambert a) at_top"
by (intro uniform_limit_lambert) (use r in auto)
show "∀⇩F n in sequentially. continuous_on (cball q r)
(λq. ∑k<n. a (Suc k) * q ^ Suc k / (1 - q ^ Suc k))" using q r
by (auto intro!: always_eventually continuous_intros simp del: power_Suc dest!: power_eq_1_iff)
qed auto
hence "continuous_on (ball q r) (lambert a)"
by (rule continuous_on_subset) auto
thus ?thesis using ‹r > 0›
by (subst (asm) continuous_on_eq_continuous_at) auto
qed
with assms show ?thesis
by (intro continuous_at_imp_continuous_on) auto
qed
lemma continuous_on_lambert' [continuous_intros]:
fixes a :: "nat ⇒ 'a :: {real_normed_field, banach, heine_borel}"
assumes "continuous_on A f" "⋀z. z ∈ A ⟹ f z ∈ eball 0 (lambert_conv_radius a) - sphere 0 1"
shows "continuous_on A (λz. lambert a (f z))"
by (rule continuous_on_compose2[OF continuous_on_lambert[OF order.refl] assms(1)])
(use assms(2) in auto)
lemma tendsto_lambert [tendsto_intros]:
fixes a :: "nat ⇒ 'a :: {real_normed_field, banach, heine_borel}"
assumes "(f ⤏ c) F" "c ∈ eball 0 (lambert_conv_radius a) - sphere 0 1"
shows "((λx. lambert a (f x)) ⤏ lambert a c) F"
proof -
have "open (eball 0 (lambert_conv_radius a) - sphere 0 1 :: 'a set)"
by (intro open_Diff closed_sphere) auto
hence "isCont (lambert a) c"
using continuous_on_lambert[OF order.refl]
by (intro continuous_on_interior) (use assms in ‹auto simp: interior_open›)
thus ?thesis
using assms(1) by (simp add: isCont_tendsto_compose)
qed
text ‹
If $\sum_{n=1}^\infty a(n)$ exists, the Lambert series of $a(n)$ tends to it for $q\to\infty$.
›
lemma tendsto_lambert_at_infinity:
assumes "summable (a :: nat ⇒ 'a :: {real_normed_field, banach, heine_borel})"
shows "(lambert a ⤏ -(∑n. a (Suc n))) at_infinity"
proof (rule Lim_transform_eventually)
have "((λq. -(∑n. a (Suc n)) - lambert a (1 / q)) ⤏ -(∑n. a (Suc n)) - lambert a 0) at_infinity"
by (rule tendsto_diff tendsto_lambert tendsto_intros tendsto_divide_0 filterlim_ident)+
(use assms in ‹auto simp: lambert_conv_radius_def›)
thus "((λq. -(∑n. a (Suc n)) - lambert a (1 / q)) ⤏ -(∑n. a (Suc n))) at_infinity"
by simp
next
have "eventually (λq :: 'a. norm q > 1) at_infinity"
by (metis dual_order.strict_trans1 eventually_at_infinity gt_ex)
thus "∀⇩F x in at_infinity. -(∑n. a (Suc n)) - lambert a (1 / x) = lambert a x"
by eventually_elim (subst lambert_reciprocal, use assms in auto)
qed
subsection ‹Power series expansion›
text ‹
By exchanging the order of summation, we can prove the power series expansion of $L(a,q)$ as
\[L(a,q) = \sum_{n=1}^\infty (a*1)(n) q^n\]
where $*$ denotes the Dirichlet product, i.e.\ $(a * 1)(n) = \sum_{d\mid n} a(d)$.
This gives particularly nice results when $a(n)$ is a number-theoretic function.
›
theorem has_sum_lambert_powser:
assumes "norm q < min 1 (conv_radius a)"
assumes "dirichlet_prod a (λ_. 1) = b"
shows "((λn. b n * q ^ n) has_sum lambert a q) {1..}"
proof -
have q: "norm q < 1" "norm q < conv_radius a"
using assms by auto
have q': "norm q < lambert_conv_radius a"
using q by (auto simp: lambert_conv_radius_def)
have "((λ(n, k). a n * q ^ (k * n)) has_sum lambert a q) ({1..} × {1..})"
proof (rule has_sum_SigmaI; (unfold prod.case)?)
show "((λn. a n * q ^ n / (1 - q ^ n)) has_sum lambert a q) {1..}"
by (intro has_sum_lambert) (use assms in ‹auto simp: lambert_conv_radius_def›)
next
show "(λ(n, k). a n * q ^ (k * n)) summable_on {1..} × {1..}"
proof (rule abs_summable_summable)
show "(λ(n, k). a n * q ^ (k * n)) abs_summable_on {1..} × {1..}"
proof (rule summable_on_SigmaI; (unfold prod.case)?)
fix n :: nat
assume n: "n ∈ {1..}"
have "((λk. norm (a n) * (norm q ^ n) ^ k) has_sum
(norm (a n) * (norm q ^ n / (1 - norm q ^ n)))) {1..}"
using has_sum_geometric[of "norm q ^ n" 1] q n
by (intro has_sum_cmult_right) (auto simp: norm_power power_less_1_iff)
thus "((λk. norm (a n * q ^ (k * n))) has_sum
(norm (a n) * norm q ^ n / (1 - norm q ^ n))) {1..}"
by (simp add: norm_mult norm_power norm_divide mult_ac flip: power_mult)
next
show "(λn. norm (a n) * norm q ^ n / (1 - norm q ^ n)) summable_on {1..}"
proof (rule abs_summable_summable)
show "(λx. norm (a x) * norm q ^ x / (1 - norm q ^ x)) abs_summable_on {1..}"
using abs_summable_on_lambert[of "of_real (norm q)" "λn. norm (a n)"] q' q
by (cases "summable (λn. norm (a n))")
(auto simp: lambert_conv_radius_def split: if_splits)
qed
qed auto
qed
next
fix n :: nat
assume n: "n ∈ {1..}"
have "((λk. a n * (q ^ n) ^ k) has_sum a n * (q ^ n / (1 - q ^ n))) {1..}"
using has_sum_geometric[of "q ^ n" 1] q n
by (intro has_sum_cmult_right) (auto simp: power_less_1_iff norm_power)
thus "((λk. a n * q ^ (k * n)) has_sum a n * q ^ n / (1 - q ^ n)) {1..}"
by (simp add: mult_ac flip: power_mult)
qed
also have "?this ⟷ ((λ(n, d). a d * q ^ n) has_sum lambert a q) (SIGMA n:{1..}. {d. d dvd n})"
by (rule has_sum_reindex_bij_witness[of _ "λ(m, d). (d, m div d)" "λ(n,k). (n * k, n)"])
(auto simp: mult_ac)
finally have 1: "((λ(n, d). a d * q ^ n) has_sum lambert a q) (SIGMA n:{1..}. {d. d dvd n})" .
have 2: "((λd. a d * q ^ n) has_sum (b n * q ^ n)) {d. d dvd n}" if n: "n > 0" for n
proof -
have "((λd. a d * q ^ n) has_sum ((∑d | d dvd n. a d) * q ^ n)) {d. d dvd n}"
by (intro has_sum_cmult_left has_sum_finite finite_divisors_nat n)
also have "(∑d | d dvd n. a d) = dirichlet_prod a (λ_. 1) n"
by (simp add: dirichlet_prod_def)
also have "… = b n"
using assms by simp
finally show ?thesis .
qed
show "((λn. b n * q ^ n) has_sum lambert a q) {1..}"
using has_sum_SigmaD[OF 1, of "λn. b n * q ^ n"] 2 by simp
qed
lemma sums_lambert_powser:
assumes "norm q < min 1 (conv_radius a)"
assumes "dirichlet_prod a (λ_. 1) = b"
shows "(λn. b n * q ^ n) sums lambert a q"
proof -
from assms(2) have [simp]: "b 0 = 0"
using dirichlet_prod_0 by blast
have "((λn. b n * q ^ n) has_sum lambert a q) {1..}"
by (rule has_sum_lambert_powser) fact+
also have "?this ⟷ ((λn. b n * q ^ n) has_sum lambert a q) UNIV"
by (intro has_sum_cong_neutral) (auto simp: not_le)
finally show ?thesis
by (rule has_sum_imp_sums)
qed
lemma conv_radius_dirichlet_prod_1_ge:
fixes a b :: "nat ⇒ 'a :: {real_normed_field, banach}"
defines "b ≡ dirichlet_prod a (λ_. 1)"
shows "conv_radius b ≥ min 1 (conv_radius a)"
proof (rule conv_radius_geI_ex)
fix r :: real
assume r: "0 < r" "r < min 1 (conv_radius a)"
have "(λn. b n * of_real r ^ n) sums lambert a (of_real r)"
using r by (intro sums_lambert_powser) (auto simp: b_def)
thus "∃z. norm z = r ∧ summable (λn. b n * z ^ n)"
using r(1) by (intro exI[of _ "of_real r"]) (auto simp: sums_iff)
qed
lemma sums_lambert_powser':
assumes "norm q < min 1 (conv_radius a)"
assumes "fds b = fds a * fds_zeta" "b 0 = 0"
shows "(λn. b n * q ^ n) sums lambert a q"
using assms(1)
proof (rule sums_lambert_powser)
have "fds_nth (fds a * fds_zeta) = dirichlet_prod a (λ_. 1)"
by (auto simp: fds_nth_mult)
also have "fds a * fds_zeta = fds b"
using assms(2) by (simp only: )
also have "fds_nth (fds b) = b"
using assms(3) by (auto simp: fun_eq_iff fds_nth_fds)
finally show "dirichlet_prod a (λ_. 1) = b" ..
qed
subsubsection ‹Divisor ‹σ› function›
text ‹
For any $q$ with $|q| < 1$ and any $\alpha\in\mathbb{C}$, we have
\[\sum_{n=1}^\infty \sigma_\alpha(n) q^n = \sum_{n=1}^\infty n^\alpha \frac{q^n}{1-q^n}\]
where $\sigma_\alpha(n)$ is the divisor ‹σ› function, i.e.
$sigma_\alpha(n) = \sum_{d \mid n} d ^{\,\alpha}$.
›
lemma divisor_sigma_powser_conv_lambert:
fixes α q :: "'a :: {nat_power_normed_field, banach}"
assumes q: "norm q < 1"
shows "(λn. divisor_sigma α n * q ^ n) sums lambert (λn. nat_power n α) q"
proof (rule sums_lambert_powser')
have "conv_radius (λn. nat_power n α) = 1"
proof (rule tendsto_imp_conv_radius_eq)
have "(λn. ereal ((real n powr (α ∙ 1)) powr (1 / real n))) ⇢ 1"
unfolding one_ereal_def by (rule tendsto_ereal) real_asymp
also have "?this ⟷ (λn. ereal (norm (nat_power n α) powr (1 / real n))) ⇢ 1"
by (intro filterlim_cong refl eventually_mono[OF eventually_gt_at_top[of 0]])
(simp add: norm_nat_power)
finally show "(λn. ereal (norm (nat_power n α) powr (1 / real n))) ⇢ 1"
by (simp add: norm_nat_power)
qed auto
thus "ereal (norm q) < min 1 (conv_radius (λn. nat_power n α))"
using q by simp
next
have "fds (divisor_sigma α) = fds_zeta * fds_shift α fds_zeta"
by (rule fds_divisor_sigma)
also have "fds_shift α fds_zeta = fds (λn. nat_power n α)"
by (simp add: fds_eq_iff)
finally show "fds (divisor_sigma α) = fds (λn. nat_power n α) * fds_zeta"
by (simp add: mult.commute)
qed auto
lemma divisor_count_powser_conv_lambert:
fixes q :: "'a :: {nat_power_normed_field, banach}"
assumes q: "norm q < 1"
shows "(λn. of_nat (divisor_count n) * q ^ n) sums lambert (λ_. 1) q"
using divisor_sigma_powser_conv_lambert[OF assms, of 0]
by (simp add: divisor_sigma_0_left)
subsubsection ‹Möbius ‹μ› function›
text ‹
For any $q$ with $|q| < 1$, we have
\[\sum_{n=1}^\infty \mu(n)\frac{q^n}{1-q^n} = q\]
where $\mu(n)$ is Möbus' ‹μ› function, which is $0$ if $n$ is not squarefree (i.e.\ contains
the same prime factor more than once) and otherwise equal to $(-1)^k$, where $k$ is the
number of prime factors of $n$.
›
lemma lambert_moebius_mu:
fixes q :: "'a :: {real_normed_field, banach}"
assumes q: "norm q < 1"
shows "lambert moebius_mu q = q"
proof -
have "(λn. indicator {1} n * q ^ n) sums lambert moebius_mu q"
proof (rule sums_lambert_powser')
have "fds moebius_mu * fds_zeta = (1 :: 'a fds)"
using fds_zeta_times_moebius_mu[where ?'a = 'a] by (simp only: mult.commute)
also have "(1 :: 'a fds) = fds (indicator {1})"
by (auto simp: fds_eq_iff fds_nth_one)
finally show "fds (indicator {1} :: nat ⇒ 'a) = fds moebius_mu * fds_zeta" ..
qed (use q in ‹auto simp: conv_radius_moebius_mu›)
also have "(λn. indicator {1} n * q ^ n) = (λn. (if n = 1 then 1 else 0) * q ^ n)"
by auto
finally have "… sums lambert moebius_mu q" .
moreover have "(λn. (if n = 1 then 1 else 0) * q ^ n) sums (q ^ 1)"
by (rule powser_sums_if)
ultimately show "lambert moebius_mu q = q"
by (simp add: sums_iff)
qed
lemma lambert_conv_radius_moebius_mu:
"lambert_conv_radius (moebius_mu :: nat ⇒ 'a :: {real_normed_field, banach}) = 1"
proof -
have "¬summable (moebius_mu :: nat ⇒ 'a)"
using not_convergent_moebius_mu[where ?'a = 'a] summable_LIMSEQ_zero
by (auto simp: convergent_def)
thus ?thesis
by (simp add: lambert_conv_radius_def conv_radius_moebius_mu)
qed
subsubsection ‹Euler's totient function ‹φ››
text ‹
For any $q$ with $|q| < 1$, we have
\[\frac{q}{(1-q)^2} = \sum_{n=1}^\infty n q^n = \sum_{n=1}^\infty \varphi(n)\frac{q^n}{1-q^n}\]
where $\varphi(n)$ is Euler's totient function, i.e.\ the number of positive integers not greater
than $n$ that are coprime to $n$.
›
lemma lambert_totient:
fixes q :: "'a :: {real_normed_field, banach}"
assumes q: "norm q < 1"
shows "lambert (λn. of_nat (totient n) :: 'a) q = q / (1 - q) ^ 2"
proof -
have "(λn. of_nat n * q ^ n) sums lambert (λn. of_nat (totient n) :: 'a) q"
proof (rule sums_lambert_powser')
show "fds (of_nat :: nat ⇒ 'a) = fds (λn. of_nat (totient n)) * fds_zeta"
by (rule fds_totient_times_zeta [symmetric])
qed (use q in ‹auto simp: conv_radius_totient›)
from sums_unique2[OF this n_powser_sums[OF q]] show ?thesis .
qed
lemma lambert_conv_radius_totient:
"lambert_conv_radius (λn. of_nat (totient n) :: 'a :: {real_normed_field, banach}) = 1"
proof -
have "¬summable (λn. of_nat (totient n) :: 'a :: {real_normed_field, banach})"
using not_convergent_totient[where ?'a = 'a] summable_LIMSEQ_zero
by (auto simp: convergent_def)
thus ?thesis
by (simp add: lambert_conv_radius_def conv_radius_totient)
qed
subsubsection ‹Mangoldt's ‹Λ› function›
text ‹
For any $q$ with $|q| < 1$, we have
\[\sum_{n=1}^\infty \ln n\ q^n = \sum_{n=1}^\infty \Lambda(n)\frac{q^n}{1-q^n}\]
where $\Lambda(n)$ is Mangoldt's function, which is defined to be equal to $\log n$ if $n$
is prime and $0$ otherwise.
›
lemma lambert_mangoldt:
fixes q :: "'a :: {real_normed_field, banach}"
assumes q: "norm q < 1"
shows "(λn. of_real (ln (Suc n)) * q ^ (Suc n)) sums lambert mangoldt q"
proof -
have "(λn. (if n = 0 then 0 else of_real (ln n)) * q ^ n) sums lambert mangoldt q"
by (rule sums_lambert_powser')
(use q fds_mangoldt_times_zeta[where ?'a = 'a] in ‹auto simp: conv_radius_mangoldt›)
also have "?this ⟷ (λn. (if Suc n = 0 then 0 else of_real (ln (Suc n))) * q ^ Suc n) sums lambert mangoldt q"
by (subst sums_Suc_iff) auto
also have "… ⟷ ?thesis"
by simp
finally show ?thesis .
qed
lemma lambert_conv_radius_mangoldt:
"lambert_conv_radius (mangoldt :: nat ⇒ 'a :: {real_normed_field, banach}) = 1"
proof -
have "¬summable (mangoldt :: nat ⇒ 'a :: {real_normed_field, banach})"
using not_convergent_mangoldt[where ?'a = 'a] summable_LIMSEQ_zero
by (auto simp: convergent_def)
thus ?thesis
by (simp add: lambert_conv_radius_def conv_radius_mangoldt)
qed
subsubsection ‹Liouville's ‹λ› function›
text ‹
For any $q$ with $|q| < 1$, we have
\[\sum_{n=1}^\infty q^{n^2} = \sum_{n=1}^\infty \lambda(n)\frac{q^n}{1-q^n}\]
where $\lambda(n)$ is Liouville's function, which is defined as the number of prime factors
of $n$ (taking multiplicity into account).
›
lemma lambert_liouville_lambda:
fixes q :: "'a :: {real_normed_field, banach}"
assumes q: "norm q < 1"
shows "(λn. ind is_square n * q ^ n) sums lambert liouville_lambda q"
by (rule sums_lambert_powser')
(use q fds_liouville_lambda_times_zeta[where ?'a = 'a]
in ‹auto simp: conv_radius_liouville_lambda mult_ac›)
lemma lambert_liouville_lambda':
fixes q :: "'a :: {real_normed_field, banach}"
assumes q: "norm q < 1"
shows "(λn. q ^ ((n+1) ^ 2)) sums lambert liouville_lambda q"
proof -
have "(λn. ind is_square ((n+1)^2) * q ^ ((n+1) ^ 2)) sums lambert liouville_lambda q ⟷
(λn. ind is_square n * q ^ n) sums lambert liouville_lambda q"
proof (rule sums_mono_reindex)
show "strict_mono (λn::nat. (n + 1)⇧2)"
by (intro strict_monoI power_strict_mono) auto
show "ind is_square n * q ^ n = 0" if "n ∉ range (λn. (n + 1)⇧2)" for n
proof (rule ccontr)
assume "ind is_square n * q ^ n ≠ 0"
hence "is_square n" "n > 0"
by (auto simp: ind_def split: if_splits)
then obtain m where "n = m ^ 2" "m > 0"
by (elim is_nth_powerE) auto
hence "n = ((m - 1) + 1) ^ 2"
by auto
with that show False by blast
qed
qed
thus ?thesis
using lambert_liouville_lambda[OF assms] by simp
qed
lemma lambert_conv_radius_liouville_lambda:
"lambert_conv_radius (liouville_lambda :: nat ⇒ 'a :: {real_normed_field, banach}) = 1"
proof -
have "¬summable (liouville_lambda :: nat ⇒ 'a)"
using not_convergent_liouville_lambda[where ?'a = 'a] summable_LIMSEQ_zero
by (auto simp: convergent_def)
thus ?thesis
by (simp add: lambert_conv_radius_def conv_radius_liouville_lambda)
qed
subsection ‹Expressing a Lambert series in terms of a power series›
text ‹
Let $a(n)$ be a sequence of numbers. Then we can express the value of the Lambert series
as an infinite sum in terms of the ``normal'' power series $f(q) = \sum_{k=1}^\infty a(k) q^k$:
\[ L(a, q) = \sum_{n=1}^\infty f(q^n) \]
The proof is quite obvious, by expanding $f(q^n)$ into its power series and then switching
the order of summation.
This gives us a number of interesting relationships, including a connection between
$L(n^a, q)$ and the polylogarithm function $\text{Li}_{-a}$.
›
theorem lambert_conv_powser_has_sum:
assumes q: "norm q < min 1 (conv_radius a)" and [simp]: "a 0 = 0"
defines "f ≡ (λq. ∑n. a n * q ^ n)"
shows "((λn. f (q ^ n)) has_sum lambert a q) {1..}"
proof -
have "((λ(k, n). a k * (q ^ k) ^ n) has_sum lambert a q) ({1..} × {1..})"
proof (rule has_sum_SigmaI; (unfold prod.case)?)
fix k :: nat
assume k: "k ∈ {1..}"
show "((λy. a k * (q ^ k) ^ y) has_sum (a k * (q ^ k / (1 - q ^ k)))) {1..}"
by (intro has_sum_cmult_right has_sum_geometric_from_1)
(use q k in ‹auto simp: power_less_1_iff norm_power›)
next
have "((λk. a k * q ^ k / (1 - q ^ k)) has_sum lambert a q) {1..}"
using q by (intro has_sum_lambert) (auto simp: lambert_conv_radius_def split: if_splits)
thus "((λk. a k * (q ^ k / (1 - q ^ k))) has_sum lambert a q) {1..}"
by (simp add: field_simps)
next
show "(λ(k, n). a k * (q ^ k) ^ n) summable_on {1..} × {1..}"
proof (rule abs_summable_summable, rule summable_on_SigmaI;
(unfold prod.case norm_divide norm_power norm_mult norm_one norm_of_nat)?)
fix k :: nat assume k: "k ∈ {1..}"
show "((λn. norm (a k) * (norm q ^ k) ^ n) has_sum
(norm (a k) * (norm q ^ k / (1 - norm q ^ k)))) {1..}"
by (intro has_sum_cmult_right has_sum_geometric_from_1)
(use k q in ‹auto simp: power_less_1_iff›)
next
show "(λn. norm (a n) * (norm q ^ n / (1 - norm q ^ n))) summable_on {1..}"
using summable_on_lambert[of "norm q" "λn. norm (a n)"] q
by (auto simp: lambert_conv_radius_def split: if_splits)
qed auto
qed
hence *: "((λ(n, k). a k * q ^ (k * n)) has_sum lambert a q) ({1..} × {1..})"
by (subst (asm) has_sum_swap) (simp_all flip: power_mult add: mult.commute)
show ?thesis
proof (rule has_sum_SigmaD [OF *]; unfold prod.case)
fix n :: nat assume n: "n ∈ {1..}"
have "ereal (norm q ^ n) ≤ ereal (norm q ^ 1)"
unfolding ereal_less_eq using n q by (intro power_decreasing) auto
also have "… < conv_radius a"
using q by (simp add: less_eq_ereal_def)
finally have norm_q_n: "norm q ^ n < conv_radius a" .
have "((λk. a k * (q ^ n) ^ k) has_sum f (q ^ n)) UNIV"
proof (rule norm_summable_imp_has_sum)
from norm_q_n show "((λk. a k * (q ^ n) ^ k) sums f (q ^ n))"
unfolding f_def by (intro summable_sums summable_in_conv_radius) (auto simp: norm_power)
next
show "summable (λk. norm (a k * (q ^ n) ^ k))"
by (rule abs_summable_in_conv_radius) (use norm_q_n in ‹auto simp: norm_power›)
qed
also have "?this ⟷ ((λk. a k * q ^ (k * n)) has_sum f (q ^ n)) {1..}"
by (intro has_sum_cong_neutral) (auto simp: mult.commute not_le simp flip: power_mult)
finally show … .
qed
qed
lemma lambert_conv_powser_has_sum':
assumes "norm q < r" and "r ≤ 1"
assumes "⋀q. norm q < r ⟹ (λn. a (Suc n) * q ^ Suc n) sums f q"
shows "((λn. f (q ^ n)) has_sum lambert a q) {1..}"
proof -
define a' where "a' = (λk. if k = 0 then 0 else a k)"
have "norm q < r"
by fact
also have "r ≤ conv_radius a"
proof (rule conv_radius_geI_ex)
fix r' assume r': "0 < r'" "ereal r' < ereal r"
show "∃x. norm x = r' ∧ summable (λn. a n * x ^ n)"
by (rule exI[of _ "of_real r'"], subst summable_Suc_iff [symmetric])
(use assms(3)[of "of_real r'"] r' in ‹simp add: sums_iff›)
qed
also have "conv_radius a = conv_radius a'"
by (intro conv_radius_cong eventually_mono[OF eventually_gt_at_top[of 0]]) (auto simp: a'_def)
finally have "((λn. (∑k. a' k * (q ^ n) ^ k)) has_sum lambert a' q) {1..}"
using assms(1,2) by (intro lambert_conv_powser_has_sum) (auto simp: a'_def)
also have "?this ⟷ ((λn. f (q ^ n)) has_sum lambert a' q) {1..}"
proof (rule has_sum_cong)
fix n :: nat assume n: "n ∈ {1..}"
have "norm q ^ n ≤ norm q ^ 1"
using assms(1,2) n by (intro power_decreasing) auto
also have "… < r"
using assms(1,2) by simp
finally have "(λk. a (Suc k) * (q ^ n) ^ Suc k) sums f (q ^ n)"
by (intro assms) (auto simp: norm_power)
hence "(λk. a' (Suc k) * (q ^ n) ^ Suc k) sums f (q ^ n)"
by (simp add: a'_def)
hence "(λk. a' k * (q ^ n) ^ k) sums f (q ^ n)"
by (subst (asm) sums_Suc_iff) (simp add: a'_def)
thus "(∑k. a' k * (q ^ n) ^ k) = f (q ^ n)"
by (simp add: sums_iff)
qed
also have "lambert a' q = lambert a q"
by (simp add: a'_def)
finally show ?thesis .
qed
lemma lambert_conv_powser_sums:
assumes q: "norm q < min 1 (conv_radius a)" and [simp]: "a 0 = 0"
defines "f ≡ (λq. ∑n. a n * q ^ n)"
shows "(λn. f (q ^ Suc n)) sums lambert a q"
proof -
have "((λn. f (q ^ n)) has_sum lambert a q) {1..}"
unfolding f_def by (rule lambert_conv_powser_has_sum) fact+
also have "?this ⟷ ((λn. f (q ^ Suc n)) has_sum lambert a q) UNIV"
by (rule has_sum_reindex_bij_witness[of _ Suc "λn. n - 1"]) auto
finally show ?thesis
by (rule has_sum_imp_sums)
qed
lemma lambert_conv_powser_sums':
assumes "norm q < r" and "r ≤ 1"
assumes "⋀q. norm q < r ⟹ (λn. a (Suc n) * q ^ Suc n) sums f q"
shows "(λn. f (q ^ Suc n)) sums lambert a q"
proof -
have "((λn. f (q ^ n)) has_sum lambert a q) {1..}"
by (rule lambert_conv_powser_has_sum') fact+
also have "?this ⟷ ((λn. f (q ^ Suc n)) has_sum lambert a q) UNIV"
by (rule has_sum_reindex_bij_witness[of _ Suc "λn. n - 1"]) auto
finally show ?thesis
by (rule has_sum_imp_sums)
qed
lemma lambert_mult_exp_conv_powser_has_sum:
assumes "norm q < r" and "r ≤ 1" and c: "norm c ≤ 1"
assumes "⋀q. norm q < r ⟹ (λn. a (Suc n) * q ^ Suc n) sums f q"
shows "((λn. f (c * q ^ n)) has_sum lambert (λn. c ^ n * a n) q) {1..}"
proof (rule lambert_conv_powser_has_sum')
fix q :: 'a assume q: "norm q < r"
have "norm c * norm q ≤ 1 * norm q"
using q c by (intro mult_right_mono) auto
hence cq: "norm c * norm q < r"
using q by simp
have "(λn. a (Suc n) * (c * q) ^ Suc n) sums f (c * q)"
by (rule assms) (use cq in ‹simp add: norm_mult›)
thus "(λn. c ^ Suc n * a (Suc n) * q ^ Suc n) sums f (c * q)"
by (simp add: power_mult_distrib algebra_simps)
qed (use assms in auto)
lemma lambert_mult_exp_conv_powser_sums:
assumes "norm q < r" and "r ≤ 1" and c: "norm c ≤ 1"
assumes "⋀q. norm q < r ⟹ (λn. a (Suc n) * q ^ Suc n) sums f q"
shows "((λn. f (c * q ^ Suc n)) sums lambert (λn. c ^ n * a n) q)"
proof (rule lambert_conv_powser_sums')
fix q :: 'a assume q: "norm q < r"
have "norm c * norm q ≤ 1 * norm q"
using q c by (intro mult_right_mono) auto
hence cq: "norm c * norm q < r"
using q by simp
have "(λn. a (Suc n) * (c * q) ^ Suc n) sums f (c * q)"
by (rule assms) (use cq in ‹simp add: norm_mult›)
thus "(λn. c ^ Suc n * a (Suc n) * q ^ Suc n) sums f (c * q)"
by (simp add: power_mult_distrib algebra_simps)
qed (use assms in auto)
lemma lambert_power_int_has_sum_polylog_gen:
fixes q :: complex
assumes q: "norm q < 1" and c: "norm c ≤ 1"
shows "((λn. polylog (-a) (c * q ^ n)) has_sum lambert (λn. c ^ n * of_nat n powi a) q) {1..}"
using q
proof (rule lambert_mult_exp_conv_powser_has_sum)
show "(λn. of_nat (Suc n) powi a * q ^ Suc n) sums polylog (-a) q"
if "norm q < 1" for q :: complex
using sums_polylog[of q "-a"] that by simp
qed (use assms in auto)
lemma has_sum_lambert_recip_complex_gen:
fixes q :: complex
assumes q: "norm q < 1" and c: "norm c ≤ 1"
shows "((λk. -ln (1 - c * q ^ k)) has_sum lambert (λn. c ^ n / of_nat n) q) {1..}"
proof -
have "((λn. polylog 1 (c * q ^ n)) has_sum lambert (λn. c ^ n * of_nat n powi - 1) q) {1..}"
using lambert_power_int_has_sum_polylog_gen[OF q, of c "-1"] q c by simp
also have "?this ⟷ ((λn::nat. -ln (1 - c * q ^ n)) has_sum
lambert (λn. c ^ n * of_nat n powi -1) q) {1..}"
proof (intro has_sum_cong)
fix n :: nat assume n: "n ∈ {1..}"
have "norm c * norm (q ^ n) ≤ 1 * norm (q ^ n)"
using q c by (intro mult_right_mono) auto
also have "norm (q ^ n) ≤ norm q ^ 1"
using n q unfolding norm_power by (intro power_decreasing) auto
also have "1 * norm q ^ 1 < 1"
using q by simp
finally have cq: "norm (c * q ^ n) < 1"
by (simp_all add: norm_mult)
thus "polylog 1 (c * q ^ n) = -ln (1 - c * q ^ n)"
by (subst polylog_1) auto
qed
also have "(λn. c ^ n * of_nat n powi -1) = (λn. c ^ n / of_nat n :: complex)"
by (auto simp: field_simps)
finally show ?thesis .
qed
lemma has_sum_lambert_recip_complex:
fixes q :: complex
assumes q: "norm q < 1"
shows "((λk. -ln (1 - q ^ k)) has_sum lambert (λn. 1 / of_nat n) q) {1..}"
using has_sum_lambert_recip_complex_gen[OF assms, of 1] by simp
lemma has_sum_lambert_recip_complex':
fixes q :: complex
assumes q: "norm q < 1"
shows "((λk. -ln (1 + q ^ k)) has_sum lambert (λn. (-1) ^ n / of_nat n) q) {1..}"
using has_sum_lambert_recip_complex_gen[OF assms, of "-1"] by simp
lemma has_sum_lambert_poly_complex:
fixes q :: complex and a :: nat
assumes q: "norm q < 1" and a: "a > 0"
defines "E ≡ poly (eulerian_poly a)"
shows "((λn. E (q ^ n) * q ^ n / (1 - q ^ n) ^ (a + 1)) has_sum
lambert (λn. complex_of_nat n ^ a) q) {1..}"
proof -
have "((λn. polylog (-a) (q ^ n)) has_sum lambert (λn. of_nat n ^ a) q) {1..}"
using lambert_power_int_has_sum_polylog_gen[OF q, of 1 a] q by simp
also have "?this ⟷ ((λn. E (q ^ n) * q ^ n / (1 - q ^ n) ^ (a+1))
has_sum lambert (λn. of_nat n ^ a) q) {1..}"
proof (rule has_sum_cong)
fix n :: nat assume n: "n ∈ {1..}"
have "norm (q ^ n) ≤ norm q ^ 1"
using n q unfolding norm_power by (intro power_decreasing) auto
also have "… < 1"
using q by simp
finally show "polylog (-a) (q ^ n) = E (q ^ n) * q ^ n / (1 - q ^ n) ^ (a+1)"
using a by (subst polylog_neg_int_left)
(auto simp: E_def power_int_diff power_int_minus divide_simps)
qed
finally show ?thesis .
qed
lemma lambert_minus1_power_has_sum:
assumes q: "norm q < 1"
shows "((λn. q ^ n / (1 + q ^ n)) has_sum lambert (λn. (-1) ^ Suc n) q) {1..}"
using q
proof (rule lambert_conv_powser_has_sum')
show "(λn. (-1) ^ Suc (Suc n) * q ^ Suc n) sums (q / (1 + q))" if "norm q < 1" for q :: 'a
proof -
have "(λn. (-1) ^ Suc (Suc n) * q ^ Suc n) sums (1 - 1 / (1 + q))"
using sums_minus[OF geometric_sums[of "-q"]] that
by (subst sums_Suc_iff) (auto simp: field_simps power_minus')
also have "1 - 1 / (1 + q) = q / (1 + q)"
using that by (auto simp: divide_simps add_eq_0_iff)
finally show ?thesis .
qed
qed auto
lemma lambert_exp_has_sum:
fixes q :: "'a :: {real_normed_field, banach}"
assumes q: "norm q < 1" and a: "norm a ≤ 1"
shows "((λn. a * q ^ n / (1 - a * q ^ n)) has_sum lambert (λn. a ^ n) q) {1..}"
proof -
have "((λn. a * q ^ n / (1 - a * q ^ n)) has_sum lambert (λn. a ^ n * 1) q) {1..}"
using q
proof (rule lambert_mult_exp_conv_powser_has_sum)
show "(λn. 1 * q ^ Suc n) sums (q / (1 - q))" if "norm q < 1" for q :: 'a
using geometric_sums_gen[of q 1] that by simp
qed (use a in auto)
thus ?thesis
by simp
qed
subsection ‹Connection to Euler's function›
text ‹
In this section, we show a connection between Lambert series and Euler's function:
\[\varphi(q) = \prod_{k=1}^\infty (1 - q ^ k)\]
(not to be confused with Euler's totient function, commonly denoted with $\varphi(n)$)
For this, we apply the results from the previous section to $a(n) = \frac{1}{n}$ to obtain:
\[\sum_{k=1}^\infty \ln (1 - q^k) = -L\big(\tfrac{1}{n}, q\big)\]
›
lemma sums_lambert_recip_complex:
fixes q :: complex
assumes q: "norm q < 1"
shows "((λk. -ln (1 - q ^ Suc k)) sums lambert (λn. 1 / of_nat n) q)"
using q
proof (rule lambert_conv_powser_sums')
show "(λk. 1 / of_nat (Suc k) * q ^ Suc k) sums - Ln (1 - q)" if "norm q < 1" for q
using sums_minus[OF Ln_series[of "-q"]] that
by (subst sums_Suc_iff) (simp_all add: power_minus')
qed auto
lemma sums_lambert_recip_complex':
fixes q :: complex
assumes q: "norm q < 1"
shows "((λk. -ln (1 + q ^ Suc k)) sums lambert (λn. (-1)^n / of_nat n) q)"
using q
proof (rule lambert_conv_powser_sums')
show "(λk. (-1)^Suc k / of_nat (Suc k) * q ^ Suc k) sums - Ln (1 + q)" if "norm q < 1" for q
using sums_minus[OF Ln_series[of "q"]] that
by (subst sums_Suc_iff) (simp_all add: power_minus')
qed auto
text ‹
By exponentiating this, we get:
\[\varphi(q) \stackrel{\text{def}}{=} \prod_{n=1}^\infty \big(1 - q^n\big) =
\exp\left(-\sum_{n=1}^\infty \frac{1}{n}\frac{q^n}{1-q^n}\right)\]
In other words, the Lambert sum $\sum \frac{1}{n}\frac{q^n}{1-q^n}$ is a logarithm of
Euler's function $\varphi(q)$.
Note that this does not show that this is ∗‹the› logarithm of $\varphi(q)$, but merely that
it is ∗‹one› of the branches of the multi-valued logarithm of $\varphi(q)$. Nevertheless, we
will -- just like is typically in textbooks -- ignore this in our informal explanations
and write $\ln \varphi(q)$.
›
theorem euler_phi_conv_lambert:
fixes q :: complex
assumes q: "norm q < 1"
shows "(λn. 1 - q ^ Suc n) has_prod exp (-lambert (λn. 1 / of_nat n) q)"
proof -
have not_1: "q ^ n ≠ 1" if "n > 0" for n
using that q by (auto dest: power_eq_1_iff)
have "(λn. exp (-ln (1 - q ^ Suc n))) has_prod exp (lambert (λn. 1 / of_nat n) q)"
by (intro sums_imp_has_prod_exp sums_lambert_recip_complex q)
also have "(λn. exp (-ln (1 - q ^ Suc n))) = (λn. inverse (1 - q ^ Suc n))"
using q unfolding exp_minus by (subst exp_Ln) (auto simp del: power_Suc simp: not_1)
finally have "(λn. inverse (inverse (1 - q ^ Suc n))) has_prod
inverse (exp (lambert (λn. 1 / complex_of_nat n) q))"
by (intro has_prod_inverse)
thus ?thesis
using q by (simp del: power_Suc add: exp_minus)
qed
text ‹
With our general results on Lambert series, we also know that $\ln \varphi(q)$ has the power
series expansion
\[\ln\varphi(q) = -\sum_{n=1}^\infty \sigma_{-1}(n) q^n =
-\sum_{n=1}^\infty \frac{\sigma_1(n)}{n} q^n\ .\]
›
lemma ln_euler_phi_powser:
fixes q :: complex
assumes q: "norm q < 1"
shows "(λn. divisor_sigma (-1) n * q ^ n) sums lambert (λn. 1 / of_nat n) q"
using divisor_sigma_powser_conv_lambert[OF q, of "-1"]
by (simp add: powr_minus divide_inverse)
lemma ln_euler_phi_powser':
fixes q :: complex
assumes q: "norm q < 1"
shows "(λn. divisor_sum n / n * q ^ n) sums lambert (λn. 1 / of_nat n) q"
using ln_euler_phi_powser[OF q]
by (simp add: divisor_sigma_minus divisor_sigma_1_left mult_ac)
text ‹
We also show the following variant of the above, also mentioned by Knopp:
›
theorem euler_phi_variant_conv_lambert:
fixes q :: complex
assumes q: "norm q < 1"
shows "(λn. 1 + q ^ Suc n) has_prod exp (-lambert (λn. (-1) ^ n / of_nat n) q)"
proof -
have not_1: "q ^ n ≠ -1" if "n > 0" for n
proof
assume "q ^ n = -1"
hence "norm q ^ n = 1"
by (simp flip: norm_power)
thus False
using that q by (auto dest: power_eq_1_iff)
qed
have "(λn. exp (-ln (1 + q ^ Suc n))) has_prod exp (lambert (λn. (-1)^n / of_nat n) q)"
by (intro sums_imp_has_prod_exp sums_lambert_recip_complex' q)
also have "(λn. exp (-ln (1 + q ^ Suc n))) = (λn. inverse (1 + q ^ Suc n))"
using q unfolding exp_minus
by (subst exp_Ln) (auto simp del: power_Suc simp: not_1 add_eq_0_iff)
finally have "(λn. inverse (inverse (1 + q ^ Suc n))) has_prod
inverse (exp (lambert (λn. (-1)^n / complex_of_nat n) q))"
by (intro has_prod_inverse)
thus ?thesis
using q by (simp del: power_Suc add: exp_minus)
qed
subsection ‹Application: Fibonacci numbers›
text ‹
Lastly, we show a connection between the Fibonacci numbers and Lambert series, namely that:
\[\sum_{n=1}^\infty \frac{1}{F_n} =
\sqrt{5}\left[L\big(1, \tfrac{1}{2}(3-\sqrt{5})\big) - L\big(1, \tfrac{1}{2}(7-3\sqrt{5})\big)\right]\]
›
lemma fib_closed_form_alt:
defines "φ ≡ (1 + sqrt 5) / 2"
shows "real (fib n) = (φ ^ n - (-1 / φ) ^ n) / sqrt 5"
proof -
have "real (fib n) = (φ ^ n - ((1 - sqrt 5) / 2) ^ n) / sqrt 5"
unfolding φ_def by (rule fib_closed_form)
also have "1 + sqrt 5 > 0"
by (intro add_pos_pos) auto
hence "(1 - sqrt 5) / 2 = -(1 / φ)"
by (simp add: φ_def field_simps)
finally show ?thesis
by simp
qed
theorem sum_inv_even_fib_conv_lambert:
defines "L ≡ lambert (λ_. 1)"
shows "((λn. 1 / real (fib (2*n))) has_sum
(sqrt 5 * (L ((3 - sqrt 5) / 2) - L ((7 - 3 * sqrt 5) / 2)))) {1..}"
proof -
define φ :: real where "φ = (1 + sqrt 5) / 2"
have "1 + sqrt 5 > 0"
by (intro add_pos_pos) auto
hence [simp]: "1 + sqrt 5 ≠ 0"
by auto
have pos: "φ > 1"
by (auto simp: φ_def intro: add_pos_pos)
have [simp]: "φ ^ k = 1 ⟷ k = 0" for k
by (auto simp: φ_def dest: power_eq_1_iff)
have "((λn. 1 * (1/φ^2)^n / (1 - (1/φ^2)^n) - 1 * (1/φ^4)^n / (1 - (1/φ^4)^n))
has_sum (L (1/φ^2) - L(1/φ^4))) {1..}"
unfolding L_def using pos
by (intro has_sum_diff has_sum_lambert) (auto simp: field_simps intro!: one_less_power)
also have "(λn. 1 * (1/φ^2)^n / (1 - (1/φ^2)^n) - 1 * (1/φ^4)^n / (1 - (1/φ^4)^n)) =
(λn. 1 / (φ ^ (2 * n) - (1 / φ) ^ (2 * n)))" using pos
by (simp add: divide_simps fun_eq_iff flip: power_mult)
(simp add: algebra_simps flip: power_add)?
also have "(… has_sum (L (1/φ^2) - L(1/φ^4))) {1..} ⟷
((λn. 1 / sqrt 5 * (1 / real (fib (2 * n)))) has_sum (L (1/φ^2) - L(1/φ^4))) {1..}"
proof (intro has_sum_cong)
fix n :: nat assume n: "n ∈ {1..}"
have "φ ^ (2 * n) - (1 / φ) ^ (2 * n) = (φ ^ (2 * n) - (-1 / φ) ^ (2 * n))"
by simp
also have "… = real (fib (2 * n)) * sqrt 5"
by (subst fib_closed_form_alt) (simp add: φ_def)
finally show "1 / (φ ^ (2 * n) - (1 / φ) ^ (2 * n)) = 1 / sqrt 5 * (1 / real (fib (2 * n)))"
by simp
qed
also have "1 / φ ^ 2 = (3 - sqrt 5) / 2"
by (simp add: φ_def power_divide power2_eq_square divide_simps) (auto simp: algebra_simps)?
also have "1 / φ ^ 4 = (7 - 3 * sqrt 5) / 2"
by (simp add: φ_def power_divide eval_nat_numeral divide_simps) (auto simp: algebra_simps)?
finally have "((λn. sqrt 5 * (1 / sqrt 5 * (1 / real (fib (2 * n))))) has_sum
sqrt 5 * (L ((3 - sqrt 5) / 2) - L ((7 - 3 * sqrt 5) / 2))) {1..}"
by (intro has_sum_cmult_right)
thus ?thesis
by simp
qed
end