Professional Documents
Culture Documents
John Gill
2010
1. Preliminaries:
Tannery's Theorem [1] provides sufficient conditions on the series-like expression S(n)
= a1 (n) + a 2 (n) + + a n (n) that it converge to the limit of the series a1 + a 2 + , when
lim a k (n) = a k for each k. In fact, the original theorem provided this result for a more
n →∞
_______________________________________________
Tannery's Theorem (series): Suppose that S(n) = a1 (n) + a 2 (n) + + a n (n) , where
lim a k (n) = a k for each k. Furthermore, assume a k (n) ≤ M k with
n →∞
∑M k <∞.
Then lim S(n) = a1 + a 2 + , convergent.
n →∞
________________________________________________
Etc. ||
Setting f k,n (z) = a k (n) + z , then one may write
S(n) = f1,n f 2,n f n,n (0) , or S(n) = f n,n f n −1,n f1,n (0) .
____________________________________________________
n
Tannery's Theorem (products): Suppose that P(n) = ∏ (1 + a k (n)) . If
k =1
∞
lim a k (n) = a k , and a k (n) ≤ M k with
n →∞
∑M k < ∞ , then lim P(n) = ∏ (1 + a k ) .
n →∞
k =1
_____________________________________________________
(n) ) = e∑
ln(1+ a k (n )) 3 1
Proof: (sketch) ∏ (1 + a k , ln(1+z) <
2
z if z <
2
Apply Tannery's Theorem for series . . . ||
However, the original theorem is less adaptable to more exotic expansions like continued
fractions:
a (n) a 2 (n) a n (n)
C(n) = 1 ,
1+ 1+ 1
which can be expressed as
a k (n)
C(n) = f1,n f 2,n f n,n (0) , with f k,n (z) = .
1+ z
a1 a 2 an 1
Example: The continued fraction → λ as n → ∞ , if a k <
1 + 1 ++ 1 4
1
Suppose lim a k (n) = a k for each k ≤ n and a k (n) < for all such terms .
n →∞ 4
a1 (n) a 2 (n) a n (n)
Then → λ as n → ∞
1 + 1 + + 1
A continuous analog of the Tannery Theorem is the following:
__________________________________________________________
_________________________________________________________
_________________________________________________________
lim ( a1 (n) + a 2 (n) + + a n (n) ) = lim a1 (n) + lim a 2 (n) + + lim a n (n)
n →∞ n →∞ n →∞ n →∞
lim t1,n t 2,n t n.n (z) = lim t1,n lim t 2,n lim t n,n (z)
n →∞ n →∞ n →∞ n →∞
or
lim t n,n t n-1,n t1,n (z) = lim t n,n lim t n −1,n lim t1,n (z)
n →∞ n →∞ n →∞ n →∞
Additional theory addresses scenarios in which the distribution of limits shown above
does not occur (e.g., the Riemann Integral) . . . making mathematical life a bit more
interesting.
. . .
2. Extending Tannery's Theorem to Inner Composition with Contractions
Theorem (Henrici [1], 1974). Let f be analytic in a simply-connected region S and continuous on the
closure S' of S. Suppose f(S') is a bounded set contained in S. Then f n (z) = f f f (z) → α , the
attractive fixed point of f in S, for all z in S'.
___________________________________________________________
This result can be extended to forward iteration (or inner composition) involving a
sequence of functions:
___________________________________________________
(Note: This result is sometimes called the Lorentzen-Gill Theorem since the second author obtained the
result in a specific case in previous papers [4], [6])
Comment: The condition lim f k,n (z) = f k (z) , if discarded, allows the possibility of
n →∞
divergence by oscillation: viz.,
Proof: Theorem 2.1 defines λ . Write Zp,n = f p +1,n f p + 2,n f n,n (z) . Then
Fn,n (z) − λ = Fp,n (Z p,n ) − λ
ε
For the second term in the inequality, choose and fix p sufficiently large that Fp (z) − λ < for all z in
2
ε
D. For the first term, choose n sufficiently large to insure Fp,n (z) − Fp (z) <
for all z in D. This is true
2
since a finite composition of functions of the type described above, converging uniformly on D, will also
converge uniformly on D. ||
A Tannery Transformation: An existing compositional structure Fn (z) = f1 f 2 f n (z) may be
transformed using these ideas:
Proof: Set g k,n (z) = f k (t k,n (z)) and apply the theorem. ||
α1 (α1 + 1) α 2 (α 2 + 1) α (α + 1)
Cn (ω) = ... n n
1 + 1 + 1 + ω
1 1
converges under the following stipulations: αn < , ω<
. The { α n } are the
5 2
α (α + 1)
attractive fixed points of the linear fractional transformations t k (ω) = k k .
1+ ω
Thus, one may write Cn (ω) = t1 t 2 t n (ω) . (If α n ≡ α , then lim Cn (ω) = α ).
n →∞
lim α k (n) = α k for each k, we have lim Cn,n (ω) = lim Cn (ω) .
n →∞ n →∞ n →∞
1 1 1
Example: Nested logarithms Ln 2 + Ln 3 + Ln ( 4 + )
2 3 4
1
Here, t k (z) = Ln(k + 1 + z) , z < 1 ⇒ t k (z) ≤ ρ < 1 . Thus,
k +1
t1 t n (z) → .438699 . Similarly,
1 1
Ln 2 ⋅ a1 (n) + Ln ( 3 ⋅ a 2 (n) + ) , where 1 ≤ a k (n) → 1 , converges to the same
2 3
limit. The obvious choice for an initial value of z is 0.
∞ n
t k (z) = ∫ ϕk (t, z)dt or t k,n (z) = ∫ ϕk (t, z)dt
0 0
n
For instance: t k,n (z) = ∫ e − t(k +1+ε+ z) dt , z < 1. Giving rise to a complicated expansion
0
that converges to the limit of the continued fraction:
1
t1,n t n,n (z) →
1
2+ε+
1
3+ε+
4+ε
. . .
Extending Tannery's Theorem to Inner Compositions without Contractions
____________________________________________________
F(z) = lim Fk (z) , and Φ p,n (z) = φ1,n φ2,n φp,n (z) , Zp,n = φp +1,n φp + 2,n φn ,n (z) .
k →∞
Then
lim Φ n,n (z) = F(z) .
n →∞
_______________________________________________________
z R
Outline of Proof: Consider z ≤ R . Then, φk,n (z) ≤ ≤ , which may
1 − ρk z 1 − ρk R
be repeated to give Zp,n ≤ R 0 = 2R . The original Tannery's Theorem shows that
φk ,n (z) → f k (z) uniformly on ( z ≤ R 0 ) = S. As in Kojima's Theorem [10],
2
ρp z
φp,n (z) − z ≤ , which may be used to prove that
1 − ρp z
∞
Zp,n − z ≤ 2R ⋅ 2
0 ∑ρ
k = p +1
k → 0 as p → ∞ .
___________________________________________________________
Proof: Set Φ p,n = t p,n t p +1,n t n,n (z) and Ψ p,n = t p t p +1 t n (z)
Then
Φ1,n − Ψ1,n ≤ t1,n (Φ 2,n ) − t1 (Φ 2,n ) + t1 (Φ 2,n ) − t1 (Ψ 2,n )
Since the values of { ρk } are not necessarily less than one, { Ψ1,n (z) } might diverge and
{ Φ1,n (z) } simply track that sequence, assuming t k,n → t k rapidly enough. ||
C k (n)z
Here t k,n (w) = , with w < r < 1 , z < R , and C k (n) < C .
1+ w
CR r(1-r)
t k,n (w) < with R < insures t k,n (w) < r when w < r .
1− r C
R R
From (a) of Theorem 2.4 , t k,n (w) − t k (w) < Ck (n) − Ck < σk (n) , and
1− r 1− r
CR
(b) t k,n (w1 ) − t k ,n (w 2 ) < = ρ . Therefore
(1-r)2
n
R
t1,n t 2,n t n,n (z) − t1 t 2 t n (z) <
1-r
∑ρ
k =1
k −1
σk (n) .
1 1 k
To illustrate, let r = , R= , C=1 , σ k (n)= 3 . Hence ρ k ≡ 1 .
2 4 n
k k
Employing Theorem 2.4, set t k,n (z) = ∫ f (x, n)dx + z ,
k −1
and t k (z) = ∫ g(x)dx + z ,
k −1
with lim F(x, n) = g(x) . Then t k,n (z) − t k (z) ≤ Max f (x, n) − g(x) = ε k (n)
n →∞ k −1≤ x ≤ k
If the sum tends to zero, the first integral grows closer to the second, even if the second
integral diverges.
n
1
Example: ∫ sin(x + σ(n))dx ,
0
where , e.g., σ(n) <
n3
, approximates to any required
n
degree of accuracy the divergent integral ∫ sin(x)dx
0
R n,n → lim σ1 + σ 2 + + σ n ?
n →∞
1 1 1
Example: 1 + + 2 + + 3 + + − 1+ 2 + 3 + → 0 .
n n n
We find that
1 n 1
t1,n t n,n (0) − t1 t n (0) < ∑ k
n k =1 2 k!
→ 0.
n
1 k
Given φ ∈ C[0,1] and a k (n) − φ( ) < ε k (n) , where
n n
∑ε
k =1
k (n) → 0 . Then
n 1
∑ a k (n) →
k =1
∫ φ(t) dt .
0
1
1 k k2 n
k2 + n 1
Example: φ(t) = t 2
⇒ φ( ) = 3 , hence
n n n
∑
k =1 n3
→ ∫ t dt =
2
3
0
z 1 kz n
Tn ,n (z) = t1,n t 2,n t n,n (z) , where t k,n (z) = + e , n ≥ 2 , z ≤ 1.
n 6
5
Hence t k,n (z) < 1 , ρk ≡ 0 , and ε1 (n) < →0 .
3n
1 1
Thus t k,n (z) → t k (z) ≡ and Theorem 7 ⇒ Tn,n (z) → .
6 6
. . .
(1) f k,n (z) − z ≤ Cρk on S2 and (2) f k,n (z) → f k (z) as n → ∞ , uniformly on S2 .
Set Zp,n = f p +1,n f p + 2,n f n,n (z) , Fp,n (z) = f1,n f 2,n f p,n (z) , and
Fn (z) = f1 f 2 f n (z) . Then there exists a function F(z) analytic on S1 such that
Sketch of proof: Theorem 2.6 [11] shows that Fn (z) → F(z) uniformly on S1 .
Next, for z ∈ S ,
Now write Fn,n (z) − F(z) ≤ Fp,n ( Z p,n ) − Fp ( Z p,n ) + Fp ( Zp,n ) − F(z)
ε
Fix p so large that the second term on the right is < . If n is sufficiently large
2
ε
the first term on the right side is < . Hence, for z ∈ S , Fn,n (z) → F(z) on S. ||
2
Theorem 3.1: (Henrici [1], 1974). Let f be analytic in a simply-connected region S and continuous on
the closure S' of S. Suppose f(S') is a bounded set contained in S. Then f n (z) = f f f (z) → α , the
attractive fixed point of f in S, for all z in S'.
______________________________________________________________
ez ez ez
Example: let G(z) = 4 8 12 … , where z ≤ 1 . We solve the continued fraction
3+ z + 3+ z + 3+ z +
e z 4n
equation G(α) = α in the following way: Set t n (ξ) = ; let g n (z) = t1 t 2 t n (0) .
3+ z + ξ
Now calculate G n (z) = g n g n −1 g1 (z) , starting with z = 1. One obtains α = .087118118
to ten decimal places after ten iterations [7].
Next, consider a sequence of functions {g k,n } dependent upon both k and n and defined on a suitable
domain D. Define G p,n (z) = g p,n g p −1,n g1,n (z) , with p ≤ n .
_______________________________________________________
z−α
Set D = (|z|<1). Let Φ(z) : = . Then Φ : D → D is analytic there, with Φ(α) = 0 and
1 − αz
Φ −1 (0) = α . Set q k,n (z) = Φ g k,n Φ −1 (z) .
(1) g k ,n ( α ) − α ≤ g k ,n ( α ) − g k ,n ( α k ,n ) + α k ,n − α
For ε > 0 , choose K and N such that k>K and n>N imply each term of the right side of (1)
ε
is less than . This is possible for the first term because the {g k,n } are uniformly bounded on D, thus
2
equicontinuous there. Hence q k,n (0) → 0 as k, n → ∞ .
Now, the existence of the compact set Ω implies g k,n (z) ≤ µ < 1 for all z in D. Thus
(2) Sup(Sup q k,n (z) = ρ < 1 exists.
k,n z <1
Since q k,n (0) → 0 as k, n → ∞ , there exists a sequence {ε k,n } such that 0 ≤ ε k,n → 0 as k, n → ∞ ,
and q k,n (0) ≤ ε K,N for all k>K and n>N. (E.g., set ε K,N = Sup q k,n (0) )
k > K,n > N
q k,n (z)
Set H k ,n (z) = . Then H k,n (z) < 1 for all |z|<1. An application of Schwartz's Lemma [3]
ρ
gives
H k,n (0) + z
H k,n (z) ≤ ≤ H k,n (0) + z .
1 + H k,n (0) ⋅ z
Therefore
Next, set Q k,n (z) = q k,n q k −1,n q1,n (z) for all k and n. Then from (2),
Q n + m,n + m (z) = q n + m,n + m (Q n + m −1,n + m (z)) ≤ q n + m,n + m (0) + ρ Q n + m −1,n + m (z) < ε n,n + ρ Q n + m −1,n + m (z)
≤ ε n,n + ρ{ q n + m −1,n + m (0) + ρ Q n + m − 2,n + m (z) }
< ε n,n + ρε n,n + ρ2 Q n + m − 2,n + m (z)
≤ ε n,n (1 + ρ) + ρ2 { q n + m − 2,n + m (0) + ρ Q n + m −3,n + m (z) }
< ε n,n (1 + ρ) + ε n ,n ρ2 + ρ3 Q n + m −3,n + m (z)
≤ ε n,n (1 + ρ + ρ2 ) + ρ3{ q n + m −3,n + m (0) + ρ Q n + m − 4,n + m (z) }
< ε n,n (1 + ρ + ρ2 + ρ3 ) + ρ4 Q n + m − 4,n + m (z)
.
.
.
ε n,n ε n,n
< + ρm −1 Q n +1,n + m (z) < + ρm
1− ρ 1− ρ
Thus, if n and m are large enough (p is large enough) both terms of the last expression can be made as
small as one wishes. Hence Q p,p (z) → 0 as p tends to infinity. It follows immediately that
G n ,n (z) → α for all z in D. It is a simple matter to extend these results to more general simply-
connected domains, D, by using appropriate Riemann Mapping Functions. ||
α1 (α1 + 1) α 2 (α 2 + 1) α (α + 1)
Cn (ω) = ... n n
1 + 1 + 1 + ω
α n (α n + 1) α n −1 (α n −1 + 1) α (α + 1)
G n (ω) = ... 1 1
1 + 1 + 1 + ω
1 1
convergent when α n < , ω < , and lim α n = α . The { α n } are the attractive
5 2 n →∞
α (α + 1)
fixed points of the linear fractional transformations t k (ω) = k k .
1+ ω
Thus, one may write G n (ω) = t n t n −1 t1 (ω) → α , as n → ∞ .
α n (n)(α n (n) + 1) α n −1 (n)(α n −1 (n) + 1) α (n)(α1 (n) + 1)
Setting G n ,n (ω) = ... 1 ,
1 + 1 + 1 + ω
where lim α k (n) = α , we have
k,n →∞
__________________________________________________________________
Theorem 3.3a: Suppose {g k,n (ζ, z)} , with k ≤ n , is a sequence of functions analytic with respect to z
on a simply-connected domain, D, for each ζ ∈ S , a second simply-connected domain, with
g k,n (S, D) ⊂ Ω , a compact subset of D, for all k and n. Let the sequence of fixed points {α k,n (ζ )} of
{g k,n } converge to α(ζ ) uniformly on S. (I.e., α k,n → α as both k and n → ∞ , with k ≤ n ). Then
G n ,n (z) = g n,n g n −1,n g1,n (z) → α(ζ ) uniformly on S × D
___________________________________________________________________
. . .
4. Extending Results for Outer Composition without Contractions
___________________________________________________________
n +p n+p-1 n+p
H n,n + p (z) − α ≤ z − α ⋅ ∏ ρk + 2 ∑ ε k ⋅ ∏ ρ j + 2ε n + p
k =n k=n j= k +1
n −1
n
n
In particular G n (z) − α = H1,n (z) − α = z − α ∏ ρk + 2∑ ε k ∏ ρ j + 2ε n
1 1 k +1
_____________________________________________________________
1 1
Comment: The simple example in which ρk = 1 − and ε k = 2 for D=(|z|<1) lies
k k
outside the context of Theorem 3.2 , and yields, after simplification,
1 2 1 1 1 2
G n (z) − α < z−α + + + + + → 0 as n → ∞
n +1 n +1 2 3 n − 1 n 2
α k (α k + 1)
Example: Let g k (z) = , with D = {z:|z|< 12 } .
1+ z
Set
2 −1 1 1
αk = − . Then α -α k = and g k (D) ⊆ D. And
2 k+4 k+4
αk 1 1
ρk = < , g k (z) → g(z) = 4 . Thus, after simplification,
1+ z 2 1+ z
1 1 1 1 2
G n (z) − α < z −α + 2 + 2 + n −1 + = E(n)
2 ( n + 3) 2 (n + 2)
n
2 2 (5) n + 4
Applying the original Tannery's Theorem to the series component shows E(n) tends to
zero as n becomes infinite.
Extending the result above to Tannery continuous compositions, we have
_____________________________________________________________
n n-1 n
G n,n (z) − α ≤ z − α ⋅ ∏ ρk +
k =1
∑ η (n) ⋅ ∏ ρ
k=1
k
j= k +1
j + ηn (n) , ηk (n) = σ k (n) + 2ε k ||
kπ kπ k
Example: Let g k (z) = sin( + z) , g k,n (z) = sin( + 3 + z) , -1 ≤ z ≤ 1
2 2 n
k k
Then σk (n) = and ρk ≡ 1 σk (n) = 3 and ρk ≡ 1 . From (4) above,
n3 n
nπ n (n − 1)π n − 1 nπ (n − 1)π 1 1
sin( + 4 + sin( + 4 + ) − sin( + sin( + ) < (1 + ) → 0
2 n 2 n 2 2 2n n
Theorem 2.7: (Gill, [11] 2011) Let {g n } be a sequence of complex functions defined
∞
on S0 =(|z| ≤ R 0 ) . Suppose there exists a sequence {ρn } such that ∑ρ
k=1
n < ∞ and
∞
g n (z) − z < Cρn if z ≤ R 0 . Set σ = C∑ ρk and R 0 = R+σ . Then, for every
1
Theorem 2.8: (Gill 2011) Suppose the functions {g k,n } are defined on S0 = (| z |≤ R 0 ) ,
∞
{ρn } are positive, σ = C∑ ρk converges, and S = (| z |≤ R) , R + σ = R 0 . Assume
k =1
(1) g k,n (z) → g k (z) uniformly on S0 and (2) g k,n (z) − z ≤ Cρk there as well.
Then for z ∈ S ,
lim G n ,n (z) = G(z) = lim G n (z) .
n →∞ n →∞
_________________________________________________________
Outline of Proof: Theorem 2.7 shows G n = G n (z) = g n g n −1 g1 (z) → G(z) and
(2) may be used repeatedly to show g p,n g p −1,n g1,n (z) ≤ R 0 .
Now, write G n,n − G ≤ G n,n − G n + G n − G , in which
n n
G n,n − G n ≤ ∑
k = p +1
G k,n − G k −1,n + ∑
k = p +1
G k − G k −1 + G p,n − G p
∞
≤ 2∑ ρk + g p,n (G p −1,n ) − g p (G p −1,n ) + g p (G p −1,n ) − g p (G p −1 )
k =p
∞
ε ε ε ε ε
< 2⋅ + + =
12 6 6 2
if p is chosen large enough to insure ∑ρ
k =p
k <
12
,
ε
and then n is large enough to guarantee each of the last two expressions is < .
6
( {g k,n } and {g k } equicontinuous on S0 , and (1) above)
ε
Thus, for large n, Gn − G < and the proof is complete. ||
2
. . .
5. Tannery Theory Potpourri . . . Trivia and Such
Comment : Consider Tn ,n (z) = t1,n t 2,n t n,n (z) where each function is of the form
t k,n (z) = a k (n) + z and lim a k (n) = a k . Then
n →∞
Tannery's original theorem covered this sort of thing, using uniform convergence
properties:
lim[a1 (n) + a 2 (n) + + a n (n)] = a1 + a 2 +
n →∞
n
k
Example 1: a k (n) = ⇒ lim Tn,n (0) = lim ∑ a k (n) = ∞
n n →∞ n →∞
k =1
n
k 1
Example 2: a k (n) =
n 2
⇒ lim
n →∞
Tn,n (0) = lim
n →∞
∑
k =1
a k (n) =
2
n
k
Example 3: a k (n) =
n3
⇒ lim
n →∞
Tn ,n (0) = lim
n →∞
∑
k =1
a k (n) = a1 + a 2 + = 0 + 0 + = 0
1
1 k n
Example 4: a k (n) = f , f ∈ C[0,1] ⇒ lim Tn,n (0) = lim ∑ a k (n) = ∫ f (x)dx
n n n →∞ n →∞
k =1 0
Simple observations arising from the preceding examples . . .
___________________________________________________________
Theorem 5.1 : Suppose lim a k (n) ≡ 0 for S(n) = a1 (n) + a 2 (n) + + a n (n) . Then
n →∞
ρ
(a) a k (n) ≥ ⇒ S(n) ≥ ρ
n
1
(b) a k (n) ≤ 1+α , α >0 ⇒ S(n) → 0
n
m
(c) a k +1 (n) ≥ ρ ⋅ a k (n) , ρ>1 , a1 (n) ≥ ⇒ S(n) ≥ m>0
ρn −1
(d) a k +1 (n) ≤ ρ ⋅ a k (n) , ρ < 1 ⇒ S(n) → 0
___________________________________________________________
{
n
Example 5: a k (n) = 0 if k<n shows that lim Tn ,n (0) = lim ∑ a k (n) may exist in the
1 if k=n n →∞ n →∞
k =1
{
Example 6: a k (n) = 1 if k<n shows that a n (n) → 0 does not imply
0 if k=n
n
lim Tn ,n (0) = lim ∑ a k (n) exists (in a finite sense).
n →∞ n →∞
k =1
λ (k)
a k (n) − a k < ε k,n ≤ , where λ(k) is a linear function of k and β > 2 .
nβ
Series, S(n) = a1 (n) − a 2 (n) + + (−1) n +1 a n (n) , in order to converge to the alternating
series, should exhibit a fairly rapid convergence of individual terms to those of the series.
Theorem 2.4 is applicable ( t k,n (z) = a k (n) + z ) in that
n
a k (n) − a k < ε k (n) with ∑ε
1
k (n) → 0 is sufficient to insure the convergence of the
n
alternating Tannery Series: S(n) − Sn < ∑ ε k (n) → 0
1
n2 2n 2 3n 2 n +1 n ⋅ n2
Example: S(n) = − + − + (−1) . Here the
1 + n 2 1 + 4n 2 1 + 9n 2 1+ n2 ⋅ n2
1 1 1
corresponding alternating series is Sn = 1 − + − + (−1) n +1
2 3 n
1 1
We find that a k (n) − a k < 2 so that S(n) - Sn < → 0.
n n
The following result is an analogue of the familiar Integral Test for series. It doesn't
provide a spectacular new perspective on the subject . . . it's merely a curiosity that could
probably be improved:
_______________________________________________________
Theorem 5.3 : Let S(n) = a1 (n) + a 2 (n) + + a n (n) , and suppose that there exists a
non-negative, bounded and differentiable function f(x,t) , defined for x ≥ 1 and t ≥ x ,
with f (k, n) = a k (n) , f t (x, t) > 0 and f x (x, t) < 0 , f (1, n) ≤ M . Then
n
lim S(n) exists if and only if lim ∫ f (x, n)dx exists
n →∞ n →∞
1
_________________________________________________________
n n
n
The hypotheses imply S(n) and ∫ f (x, n)dx
1
are monotonic increasing. If S(n)
Example:
n n n n 1 1 π2
S(n) = + + + = (1 + + + ) →
(n + 1)12 (n + 1)2 2 (n + 1)n 2 n + 1 4 n2 6
n
n 1 n π2
n + 1 ∫1 x 2
Here, dx + → 1+1=2 ≥
n +1 6
n
n n n 1
Example: S(n) = 2
+
n +1+1 n +1+ 2 2
+ +
n +1+ n2
(≥ ∑ 2+k
k=1
→ ∞)
n
n n n 1
Here, ∫ n +1+ x
1
2
dx = Arc tan
n +1 n +1
− Arc tan → ∞,
n +1
Theorem 5.4 : Let S(n) = a1 (n) + a 2 (n) + + a n (n) , and suppose there exists a non-
negative, bounded, and differentiable function f(x,t) defined for x ≥ 0 and t ≥ x , with
f (k, n) = a k (n) . Define φ(x) = f (x, x). Suppose
f x (x, t) ≥ 0 , f t (x, t) < 0 , and φ(x) → 0 as x → ∞. Also stipulate f (1, t) → 0 as t → ∞ .
Then
n
0 ← a1 (n) ≤ S(n) − ∫ f (x, n)dx
1
≤ a n (n) → 0 as n becomes infinite.
_________________________________________________________
x 1 −2k 1
Example: f (x, t) = 2
, then f x (x, t) = 2 > 0 , f t (x, t) = 3 < 0 and φ(x) = → 0.
t t t x
1 1 1 1
The theorem shows that 1 − 2 + - S(n) → 0, or lim S(n) = , as is easily
2 n 2 n →∞ 2
1
verified by evaluating S(n) directly: S(n) = 2 (1 + 2 + 3 + + n ) .
n
Example: A slightly more sophisticated example is the following:
1 2 n
S(n) = 2 2
+ 2 2
+ + 2
1 +n 2 +n n + n2
x
Here f (x, t) = and the conditions of the theorem are satisfied for the relevant
x + t2
2
1 2 1 1 1
Ln 2 − + 2 + - S(n) → 0 , or S(n) → Ln(2)
2 n n 2n 2
The convergence is very slow. (elementary techniques also show this result)
1 2 3 n
Example: S(n) = + + + +
4 4 4
1+ n 4+n 9+n n + n4
2
n −1 n
x 1 k
Hence ∫ dx → = lim ∑
0 x2 + n4 2 n →∞
k =1 k2 + n4
_______________________________________________________
Theorem 5.5 : Let S(n) = a1 (n) + a 2 (n) + + a n (n) , and suppose there exists a non-
negative, bounded, and differentiable function f(x,t) defined for x ≥ 0 and t ≥ x , with
f (k, n) = a k (n) . Define φ(x) = f (x, x). Suppose
f x (x, t) < 0 , f t (x, t) < 0 , and φ(x) → 0 as x → ∞, and f(1,t) → 0 as t → ∞
Then
n
0 ← a n (n) ≤ S(n) − ∫ f (x, n)dx
1
≤ a1 (n) → 0 as n becomes infinite.
_________________________________________________________
Example:
n n n
S(n) = 2 2
+ 2 2
+ ... + 2
1 +n 2 +n n + n2
π
The conditions of Theorem 4 are satisfied, giving lim S(n) = .
n →∞ 4
(elementary techniques also show this result).
k
n 2 if n even 1
Example: Set a k (n) = . Then, for n even S(n) → , but for
-k 2
2 if n odd
n
1
n odd S(n) → − , although the series converges absolutely to the value ½. Here
2
a k (n) → 0 as n → ∞ for each k . Observe that were the original Tannery Theorem
applicable, we would have a very uninteresting S(n) → 0 . However, the absolute
convergence of S(n) does imply the existence of at least one subsequence {n } with
j
{S } converging, since
nj S(n) ≤ a1 (n) + + a n (n) ≤ M .
Analytic Functions . . .
__________________________________________________________
Theorem 5.6 : Define Fn (z) = a1,n (z) + a 2,n (z) + + a n,n (z) where each term in the
Tannery series is analytic on D=(|z|<1) and there exists a positive M such that
M
a k,n (z) ≤ for each k ≤ n . Furthermore, assume there exists an interval (a,b) in
n
(-1,1) so that for each x in this interval, Fn (x) → λ (x) . Then Fn (z) → λ (z) , analytic
on D, uniformly on compact subsets of D.
___________________________________________________________
Proof: Follows immediately from the Stieltjes-Vitali Theorem [9], with Fn (z) < M . The
domain D may of course be generalized. ||
1
Corollary: Set a k,n (z) = Φ k,n (z) where each Φ k,n is analytic on D and Φ k,n (z) < M .
n
Then Fn (z) → λ (z) , analytic on D, uniformly on compact subsets of D.
1 1+nz 1 n n+ z
Example: Define Fn (z) = e + + e with D=(|z|<1). It is easily shown that
n n
10
a k,n (z) < so that Fn (z) < 10.
n
1 x+u
Writing f (x, t) = e t : f x ≥ 0 , f t < 0 , f(x,x) → 0 and f(1,t) → 0 , so that Theorem 8
t
1 x+u
n
applies , with lim Sn (u) = lim ∫ e n dx = e − 1 . Therefore, Fn (z) → λ (z) = e – 1.
n →∞ n →∞ n
1
______________________________________________________________
Theorem 5.7 : Let D=(|z|<1) and S be a simply-connected domain. Suppose there exists
a sequence of functions {f k,n (ζ, z)}k ≤ n analytic on both D and S, with
f k,n (S, D) ⊂ Ω , compact, ⊂ D . Suppose f k,n → f k uniformly on S × D as n → ∞ .
F (ζ, z) = f (ζ, z) , F (ζ, z) = F (ζ, f (ζ, z)) and
Set (1) F1 (ζ, z) =1f (ζ, z) n, F (ζ, z)n=−1 F n (ζ, f (ζ, z))
1,n 1,n p,n p −1,n p,n
ζ
Set f k,n (ζ, z) = , where C ≥ 3, 0 ≤ δ(k,n) ≤ 1, lim δ(k,n)=0
C + δ(k, n) + z n →∞
1
Let S = D = (| z |< 1) and Ω = (| z |≤ 34 ) . Thus f k,n − f k < δ(k, n) ⋅ .
(C − 1) 2
ζ ζ ζ
Then Fn,n (ζ, z) = and
C + δ(1, n) + C + δ(2, n) + + C + δ(n, n) + z
ζ ζ
lim Fn,n (ζ, z) =
n →∞ C + C +
= α (ζ ) =
1
2
(C 2 + 4ζ − C . )
Theorem 5.8 : Let f be a function analytic on D = (|z|<1), with f(0) = 0 , and |f(z)|<R.
Let α = α(k, n) and β=β(k,n) , k ≤ n, be real and imaginary parts of points within D.
n n n
If ∑ α + ∑ β ≤ M for all n, then
k =1 k =1
S(n) = ∑ f (α(k, n) + iβ(k, n))
k =1
≤ MR and there
n n
Corollary: Suppose ∑ α + ∑ β → 0 as n → ∞ . Then, for any function f defined as
k =1 k =1
n
above, ∑ f (α(k, n) + iβ(k, n))
k =1
→ 0 as n → ∞ .
1
n
1 k
φn = ∑ φ → ∫ φ ( x)dx .
k =1 n n 0
Then
( k ) + ( k n ) + 1 + 1n ≈ φ k
ψ ( k , n) = n
2
where φ (x) = x 2 + x + 1 , and one might
1 + ( k n ) ⋅ 1n n
11
suspect that ψ n → . Indeed, this is the case, as is seen in the following simple
6
theorem:
_______________________________________________________________________
1 n
Theorem 5.9 : Given ψ n = ∑ψ (k , n) ,
n k =1
suppose there exists an integrable function φ ( x) with
1
k
I = ∫ φ ( x)dx and ∆ ( k , n) = φ −ψ ( k , n) ≤ ε n ⋅ R , ε n → 0 .
0 n
Then limψ n = I .
n →∞
________________________________________________________________________
R
Proof: φn − ψ n ≤ ε n ⋅ n ⋅ →0 ||
n
Example: ψ n = ∑
n Sin ( )
k 2 +k
n 2 +1 k2 k2 + k
. Thus ∆(k , n) = Cos (ηk ,n ) ⋅ 2 − 2
1
≤ ≤ ⋅2 .
k =1 n n n +1 n
1
Hence limψ n = ∫ Sin( x 2 )dx .
n →∞
0
Here’s a curiosity couched as a problem: Partition the interval [0,1] into uneven,
increasingly lengthy subintervals, going from 0 to 1, to obtain
2 2
1 n n
2k 3 ( k + 1) + 2kn 2 ( n + 1)
∫ (x
2
+ 1)dx = lim ∑ Tn = lim ∑ 3
. Then manipulate Tn so as to
0
n →∞
k =1
n →∞
k =1 n3 ( n + 1)
n 1
obtain lim ∑ Tn = ∫ ( 2 x5 + 2 x )dx !
n →∞
k =1 0
References:
[1] J. Tannery, Introduction a la Theorie des Fonctions d'une Variable, Sec.183, 1886
[2] T. Bromwich, Introduction to the Theory of Infinite Series, 2nd Ed, 1926
[3] P. Henrici, Applied & Computational Complex Analysis, Vol. 1, 1974
[4] J. Gill, Compositions of Analytic Functions of the Form Fn(z) = Fn-1(fn(z)) , Lim fn(z)=f(z), J.
Comp. & Appl. Math. 23 (1988) 179-184
[5] L. Lorentzen, Compositions of Contractions, J. Comp. & Appl. Math. 32 (1990) 169-178
[6] J. Gill, Limit Periodic Iteration, Appl. Numer. Math. 4 (1988)297-308
[7] J. Gill, The Use of the Sequence Fn (z) = f n f1 (z) in Computing Fixed Points of Continued
Fractions, Products, and Series, Appl. Numer. Math. 8 (1991) 469-4
[8] J. Gill, A Tannery Transformation of Continued Fractions and Other Expansions, Comm. Anal. Th.
Cont. Fraction Vol. 1, (1992)
[9] R. Remmert, Classical topics in complex function theory, Vol. 172, Springer (1998)
[10] S. Kojima, Convergence of Infinite Compositions of Entire Functions, arXiv:1009.2833v1]
[11] J. Gill, Convergence of Infinite Comp. of Anal. Functions, www.johngill.net (2010)