Professional Documents
Culture Documents
1
Chapter 1
Series
2
Legends
\One of the great mathemati
al
hallenges of the early 18th
entury was to nd
an expression for the sum of re
ipro
al squares
1 1 1 1
(1.0.1) 1+ + + + + :::
22 32 42 52
Joh. Bernoulli eagerly sought for this expression for many de
ades." (E. Hairer,
G. Wanner, Analysis by Its Hystory, Undergraduate Texts in Mathemati
s,
Springer, 1997)
In 1689 Ja
. Bernoulli proved
onvergen
e of the series. In 1728-1729 Gold-
bah and D. Bernoulli evaluated the series with a
ura
y 0:01. Stirling in 1730
found 8 digits of the sum.
L. Euler in 1734
al
ulated the rst 18 digits (!) of the sum (1.0.1) after the
de
imal point and re
ognized 2 =6, whi
h has the same 18 digits. He
onje
tured
that the innite sum is equal to 2 =6. In 1735 Euler dis
overed an expansion
of the sine fun
tion into an innite produ
t of polynomials:
sin x x2 x2 x2 x2
(1.0.2) = 1 1 1 1 :::
x 2 222 32 2 42 2
Comparing this presentation with the standard sine series expansion
x3 x5 x7 x9
(1.0.3) sin x = x + + :::
3! 5! 7! 9!
Euler not only proved that the sum (1.0.1) is equal to 2 =6, moreover he
al
u-
lated all sums of the type
1 1 1 1
(1.0.4) 1+ k + k + k + k +:::
2 3 4 5
for even k.
Putting x = 2 in (1.0.2) he got the beautiful Wallis' Produ
t:
22446688
(1.0.5) = :::
2 13355779
whi
h had been known sin
e 1655. But Euler's rst proof of (1.0.2) was not
satisfa
tory. In 1748, in his famous \Introdu
io in Analysin Innitorum", he
presented a proof whi
h was suÆ
iently rigorous for the 18th
entury. The
series of re
ipro
al squares was named the Euler series .
If somebody wants to understand all the details of the above legend he had to
study a lot of things, up to
omplex
ontour integrals. This is why the detailed
mathemati
al exposition of Euler's series legend turns into an entire
ourse of
Cal
ulus. The fas
inating history of Euler's series is the guiding thread of the
present
ourse, \In Euler's footsteps".
3
1.1 Autore
ursion of Innite Expressions
On the
ontents of the le
ture. The le
ture presents a romanti
style
of early analyti
s. The motto of the le
ture
ould be \innity, equality and
no denitions!". \Innity" is the main personage we will play with today. We
demonstrate how innite expressions (i.e innite sums, produ
ts, fra
tions) arise
in solutions of simple equations, how it is possible to
al
ulate them, and how
the results of su
h
al
ulations apply to nite mathemati
s. In parti
ular, we
will dedu
e the Euler-Binet formula for Fibona
i numbers, the rst Euler's
formula of the
ourse. We be
ome a
quainted with the geometri
series and the
golden se
tion.
A
hilles and the turtle. The an
ient Greek philosopher Zeno
laimed that
A
hilles pursuing a turtle
ould never pass it by, in spite of the fa
t that his
velo
ity was mu
h greater than the velo
ity of the turtle. His arguments adopted
to our purposes are the following.
First Zeno proposed a pursuing algorithm for A
hilles:
Initialization. Assign to the variable \goal" the original position of the
turtle.
A
tion. Rea
h the \goal".
Corre
tion. If the
urrent turtle's position is \goal", then STOP, else
reassign to the variable \goal" the
urrent position of the turtle and go to
A
tion.
Se
ondly, Zeno remarks that this algorithms never stops if the turtle
on-
stantly moves in one dire
tion.
And nally, he notes that A
hilles has to follow his algorithm if he want
pass the turtle by. He may be not aware of this algorithm, but un
ons
iously
he must perform it. Be
ause he
annot run the turtle down without rea
hing
the original position of the turtle and then all positions of the turtle whi
h the
variable \goal" takes.
Zeno's algorithm generates a sequen
e of times ftk g, where tk is the time of
exe
ution of the k-th A
tion of the algorithm. And the whole time of work of
1
P
the algorithm is the innite sum tk ; and this sum expresses the time A
hilles
k=1
needs to run the turtle down. (The Corre
tions take zero time, be
ause A
hilles
really does not think about them.) Let us name this sum Zeno series .
Assume that both A
hilles and the turtle run with
onstant velo
ities v and
w, respe
tively. Denote the initial distan
e between A
hilles and the turtle by d0 .
Then t1 = dv0 . The turtle in this time moves by the distan
e d1 = t1 w = wv d0 .
By his se
ond A
tion A
hilles over
omes this distan
e in time t2 = dv1 = wv t1 ,
while the turtle moves away by the distan
e d2 = t2 w = wv d1 . So we see that
sequen
es of the times ftk g and distan
es fdk g satisfy the following re
urren
e
relations: tk = wv tk 1 , dk = wv dk 1 .
Hen
e ftk g as well as fdk g are geometri
progressions with ratio wv . And
the time t whi
h A
hilles needs to run the turtle down is
w w2 w w2
t = t1 + t2 + t3 + = t1 + t1 + 2 t1 + = t1 1 + + 2 + : : :
v v v v
In spite of Zeno, we know that A
hilles does
at
h up with the turtle. And
one easily gets the time t he needs to do it by the following argument: the
4
distan
e between A
hilles and the turtle permanently de
reases with the velo
ity
v w. Consequently it be
omes 0 in the time t = vd0w = t1 v vw . Comparing
the results we
ome to the following
on
lusion
v w w2 w3
(1.1.1) = 1+ + 2 + 3 +:::
v w v v v
Innite substitution. We see that some innite expressions represent nite
values. The fra
tion in the left-hand side of (1.1.1) expands into the innite
series on the right-hand side. Innite expressions play a key r^ole in mathe-
mati
s and physi
s. Solutions of equations quite often are presented as innite
expressions.
For example let us
onsider the following simple equation
(1.1.2) t = 1 + qt
Substituting on the right-hand side 1 + qt instead of t, one gets a new equation
t = 1 + q(1 + qt) = 1 + q + q2 t. Any solution of the original equation satises
this one. Repeating this tri
k, one gets t = 1+ q(1+ q(1+ qt)) = 1+ q + q2 + q3t.
Repeating this innitely many times, one eliminates t on the right hand side
and gets a solution of (1.1.2) in an innite form
1
X
(1.1.3) t = 1 + q + q2 + q3 + = qk
k=0
On the other hand, equation (1.1.2) solved in usual way gives t = 1
1 q. As a
result, we obtain the following formula
1 X1
(1.1.4) = 1 + q + q2 + q3 + q4 + = qk
1 q k=0
whi
h represents a spe
ial
ase of (1.1.1) for v = 1, w = q.
5
These two formulas are all one needs to nd the sum of geometri
series
1 k
P P1 k
q . To be exa
t, the multipli
ation formula gives the equality q =
k=0 k=1
P1
q qk . Hen
e the shift formula turns into equation x = 1 + qx, where x is
1k=0k
P
q . The solution of this equation gives us the formula (1.1.4) for the sum of
k=0
the geometri
series again.
From this formula, one
an dedu
e the formula for the sum of a nite geo-
n
P
metri
progression. By ak is denoted the sum a0 + a1 + a2 + + an . One
k=0
has
nX1 X1 X1 1 qn 1 qn
(1.1.7) qk = qk qk = =
k=0 k=0 k=n 1 q 1 q 1 q
This is an important formula whi
h was traditionally studied in s
hool.
1
P P1 k
The series kxk . To nd the sum of kx we have to apply additionally
k=0 k=1
the following addition formula ,
X1 1
X X1
(1.1.8) (ak + bk ) = ak + bk
k=1 k=1 k=1
whi
h is the last general formula for series we introdu
e in the rst le
ture.
P1 k P1
Reindexing the shift kx we give it the form (k + 1)xk+1 . Further it
k=2 k=1
splits into two parts
X1 X1 X1 X1 x
x (k + 1)x = x kx + x x = x kxk + x
k k k
k=1 k=1 k=1 k=1 1 x
by the addition formula. The rst summand is the original sum multiplied
by x. The se
ond is a geometri
series. We already know its sum. Now the
shift formula for the sum s(x) of the original series turns into equation s(x) =
x + x 1 x x + xs(x). Its solution is s(x) = (1 xx)2 :
6
Multiplying both sides of above equation by x and
olle
ting all terms
ontaining
(x) on the right-hand side, one gets x = (x) x(x) x2 (x) = x. It leads
to x
(x) =
1 x x2
p
The roots of the
p equation 1 x x2 = 0 pare 12 5 . More famous is the pair of
their inverses 12 5 . The number = 1+2 5 is so
alled golden se
tion or golden
mean. It plays
p signi
ant role in mathemati
s, ar
hite
ture and biology. Its dual
is ^ = 1 2 5 . Then ^ = 1, and +^ = 1. Hen
e (1 x)(1 x^) = 1 x x2 ,
whi
h in its turn leads to the following de
omposition:
x 1 1 1
x2 + x 1
= p ^
5 1 x 1 x
We expand both fra
tions on the right hand side into geometri
series:
1 X1 1 X1
(1.1.11) = k x k ^ = ^k xk
1 x k=0 1 x k=0
This gives the following representation for the generating fun
tion
1 X 1
(x) = p (k ^k )xk
5 k=0
On the other hand the
oeÆ
ient at xk in the original presentation of (x) is
k . Hen
e
p p
1 ( 5 + 1)k + ( 1)k ( 5 1)k
(1.1.12) k = p (k ^k ) = p
5 2k 5
This is
alled the Euler{Binet formula. It is possible to
he
k it for small k and
then prove it by indu
tion using Fibona
i re
urren
e.
7
To be pre
ise,
onsider the sequen
e
1 1 1
(1.1.14) 1; 1+ ; 1+ ; 1+ :::
1 1 1
1+ 1+
1 1
1+
1
of the so-
alled
onvergents of the
ontinued fra
tion (1.1.13). Let us remark
that all odd
onvergent are less than and all even are greater than . To see
this,
ompare the nth
onvergent with the
orresponding term of the following
sequen
e of fra
tions:
1 1 1 1
(1.1.15) 1+ ; 1+ ; 1+ ; 1+ :::
x 1 + x1 1 1
1+ 1+
1 + x1 1
1+
1 + x1
We know that for x = all terms of the above sequen
e are equal to . Hen
e
all we need is to observe how the removal of x1 ae
ts the value of the
onsidered
fra
tion. The value of the rst fra
tion of the sequen
e de
reases, the value of
the se
ond fra
tion in
reases. If we denote the value of nth fra
tion by fn , then
the value of the next fra
tion is given by the following re
urren
e relation:
1
(1.1.16) fn+1 = 1 +
fn
Hen
e the in
reasing of fn de
reases fn+1 and de
reasing of fn in
reases fn+1 .
Consequently in general all odd fra
tions of the sequen
e (1.1.15) are less than
orresponding
onvergent, and all even are greater. The re
urren
e relation
(1.1.16) is valid for the golden mean
onvergent. By this re
urren
e relation one
an qui
kly
al
ulate the rst ten
onvergent
3 5 8 13 21 34 55 89
1; 2; ; ; ; ; ; ; ; :
2 3 5 8 13 21 34 55
The golden mean lies between last two fra
tions, whi
h have the dieren
e
1 . This allows us to determine the rst four de
imal digits after the de
imal
3455 p
point of it and of 5.
Problems.
1 22k
P
1. Evaluate 33k .
k=0
2. Evaluate 1 1 + 1 1 + : : :
3. Evaluate 1 + 1 1 1 + 1 + 1 1 1 + : : :
P1 k
4. Evaluate 3k
k=1
1 k2
P
5. Evaluate 2k
k=1
8
6. De
ompose into power series the fra
tion a+1 x
7. Find the generating fun
tion for the sequen
e f2k g.
1
P
8. Find sum the k 3 k
k=1
9. Prove by indu
tion the Euler{Binet formula.
10. Evaluate 1 2 + 1 + 1 2 + 1 : : :
p
11. Approximate 2 by a rational with pre
ision 0:0001
1
12. Find the value of 1 +
1
2+
1
1+
2+:::
q p p
13. Find the value of 2+ 2+ 2+:::
By innite substitution, solve the equation x2 2x 1 = 0, and represent
14. p
2 by a
ontinued fra
tion.
15. Find the value of innite produ
t
1 1 1
2 22 24 28 : : :
16. Find a formula for n-th term of re
urrent sequen
e xn+1 = 2xn + xn 1 ,
x0 = x1 = 1.
1
P
17. Find the sum of the Fibona
i numbers k
k=1
18. Find sum 1 + 0 1 + 1 + 0 1 : : :
19. De
ompose into the sum of partial fra
tions x2 13x+2
9
1.2 Positive Series
On the
ontents of the le
ture. Innity is pregnant with paradoxes. Para-
doxes throw us down from heavens to the earth. We leave the poetry for prose,
and rationalize the \innity and equality" by working with \niteness and in-
equality". We shall lay a solid foundation for a summation theory for positive
1 1 2
P
series. And the reader will nd out what k2 = 6 pre
isely means.
k=1
1 k
P
Divergent series paradox. Let us
onsider the series 2. This is a geo-
k=0
metri
series. We know how to sum it up by autore
ursion. The autore
ursion
equation is s = 1 + 2s. The only number satisfying this equation is 1. The
sum of positive numbers turns to be negative!? Something is wrong!
A way to save the situation is to admit innity as a feasible solution. Innity
P1 k
is an obvious solution of s = 1 + 2s. The sum of any geometri
series q
k=0
with denominator q 1 is obviously innite, isn't it?
Indeed, this sum is greater than 1 + 1 + 1 + 1 + : : : , whi
h symbolizes the
innity. (The autore
ursion equation for 1 + 1 + 1 + : : : is s = s + 1. Innity is
the unique solution of this equation.)
P1 k
The series 2 represents Zeno's series in the
ase of Mighty Turtle, whi
h
k=0
is faster than A
hilles. To be pre
ise, this series arises if v = d0 = 1 and w = 2.
As the velo
ity of the turtle is greater than the velo
ity of A
hilles he never
rea
hes it. So the innity is right answer for this problem. But the negative
solution 1 also makes sense. One
ould interpret it as an event in the past.
Just the point in time when the turtle passed A
hilles.
10
denote this sum by s then by shift formulas one gets
s = 1+0 1+1+0 1+1+0 1+1+0 1+:::
s 1 = 0 1+1+0 1+1+0 1+1+0 1+1+:::
s 1 0 = 1+1+0 1+1+0 1+1+0 1+1+0+:::
Summing the numbers
olumn-wise (i.e., by the Term-wise Addition Formula),
we get
s + (s 1) + (s 1 0) = (1 + 0 1) + (0 1 + 1) + ( 1 + 1 + 0)+
+ (1 + 0 1) + (0 1 + 1) + ( 1 + 1 + 0) + : : :
The left-hand side is 3s 2. The right-hand side is the zero series. That is why
s = 32 .
The series 1 1 + 1 1 + : : : arises as Zeno's series in the
ase of a blind
A
hilles dire
ted by a
ruel Zeno, who is interested, as always, only in proving
his
laim, and a foolish, but mer
iful turtle. The blind A
hilles is not fast, his
velo
ity equals the velo
ity of the turtle. At the rst moment Zeno tells the
blind A
hilles where the turtle is. A
hilles starts the rally. But the mer
iful
turtle wishing to help him goes towards him instead of running away. A
hilles
meets the turtle half-way. But he misses it, being busy to perform the rst
step of the algorithm. When he a
omplishes this step, Zeno orders: \Turn
about!" and surprises A
hilles by saying that the turtle is on A
hilles initial
position. The turtle dis
overs that A
hilles turns about and does the same.
The situation repeats ad innitum. Now we see that assigning the sum 12 to the
series 1 1 + 1 1 + : : : makes sense. It predi
ts a
urately the time of the rst
meeting of A
hilles and turtle.
Positivity. The paradoxes dis
ussed above are dis
ouraging. Our intuition
based on handling nite sums fails when we turn to innite ones. Observe that
all paradoxes above involve negative numbers. And to eliminate the evil in its
root, let us
onsider only nonnegative numbers.
We return to the an
ient Greeks. They simply did not know what a negative
number is. But in
ontrast to Greeks, we will retain zero. A series with nonneg-
ative terms will be
alled a positive series. We will show that for positive series
all familiar laws, in
luding asso
iativity and
ommutativity, hold true and zero
terms do not ae
t the sum.
11
2 2
from 6 . Furthermore, any number
whi
h is less than 6 has to be surpassed
2 2
by partial sums eventually, when they approa
h 6
loser than by 6
. Hen
e
the ultimate sum majorizes all partial ones, and any lesser number does not.
This means that the ultimate sum is the smallest number whi
h majorizes all
partial sums.
12
But these inequalities are true for all N and M , as the following
omputations
show.
XM N
X MX+N MX +N MX
+N 1
X
bk + ak bk + ak = (ak + bk ) (ak + bk )
k=1 k=1 k=1 k=1 k=1 k=1
In the opposite dire
tion, we see that any partial sum on the right-hand
n
P Pn n
P
side (ak + bk ) splits into ak + bk . And by virtue of the One-for-All
k=1 k=1
P1 k=1P 1
prin
iple, this does not ex
eed ak + bk . Now, the All-for-One prin
iple
k=1 k=1
provides the inequality in the opposite dire
tion.
Shift Formula. The Shift Formula immediately follows from the Term-Wise
Addition. To be pre
ise, immediately from the Denition, one gets the following:
P1
a0 + 0 + 0 + 0 + 0 + = a0 and that 0 + a1 + a2 + a3 + = ak . Term-Wise
k=1
Addition of these series gives
X1 1
X
a0 + ak = (a0 + 0) + (0 + a1 ) + (0 + a2 ) + (0 + a3 ) + = ak
k=1 k=0
1
P 1
P
Term-Wise Multipli
ation Formula. ak = ak . For any partial
k=1 k=1
Pn n
P 1
P
sum from the right-hand side one has ak = ak ak by the
k=1 k=1 k=1
Distributivity Law for nite sums and One-for-All. This implies the inequality
P1 P1
ak ak by All-for-One. The opposite inequality is equivalent to
k=1 k=1
1
P P1 Pn Pn
ak 1 ak . As any partial sum ak is equal to 1 ak , whi
h does
k=1 k=1 k=1 k=1
1
P
not ex
eed 1 ak , one gets the opposite inequality.
k=1
13
1 k
P
for q is
onstru
ted in usual way by Shift and Term-Wise multipli
ation.
k=0
P1 k
It leaves only two possibilities for q , either q 1 1 or 1. For q < 1 we have
k=0
proved
onvergen
e, and for q 1 innity is the true answer.
Let us pay spe
ial attention to the
ase q = 0. We adopt a
ommon
onven-
tion:
(1.2.2) 00 = 1:
1 k
P
This means that the series 0 satises the
ommon formula for a
onvergent
k=0
P1 k
geometri
series 0 = 1 1 0 = 1. Finally we state the theorem, whi
h essen-
k=0
tially due to Eudoxus, who proved the
onvergen
e of the geometri
series with
ratio q < 1.
Theorem 1 (Eudoxus). For every nonnegative q one has
X1 1 X1
qk = for q < 1; and qk = 1 for q 1:
k=0 1 q k=0
Comparison of series Quite often exa
t summation of series is too diÆ
ult,
and for pra
ti
al purposes it is enough to know the sum approximatively. In
this
ase one usually
ompares the series with another one whi
h sum is known.
Su
h
omparison is based on the following Term-Wise Comparison Prin
iple ,
whi
h immediately follows from the Denition of Sum.
P1 P1
Prin
iple (Term-Wise Comparison). If ak bk for k, then ak bk
k=1 k=1
The only series we have so far to
ompare with are the geometri
ones. The
following lemma is very useful for this purposes.
Lemma 2 (Ratio Test). If ak+1 qak holds for some q < 1 and all terms of
P1 1
P
series ak than ak 1a0q
k=0 k=0
Proof. By indu
tion one proves inequality ak a0 qk . Now by Term-Wise Com-
1
P 1
P
parison one estimates ak from above by the geometri
series a0 q k =
a0 k=0 k=0
1 q
If the series under
onsideration satises an autore
ursion equation, to prove
its
onvergen
e usually mean to evaluate it exa
tly. For proving of
onvergen
e
the Term-Wise Comparison Prin
iple
an be strengthened. Let us say that the
1
P 1
P
series ak is eventually majorized by the series bk , if the inequality bk ak
k=1 k=1
holds for ea
h k starting from k = n for some n. The following lemma is very
useful to prove
onvergen
e.
P1
Prin
iple (Eventual Comparison). A series ak , whi
h is eventually ma-
k=1
1
P
jorized by a
onvergent series bk , is
onvergent.
k=1
14
1
P 1
P P1
Proof. Consider a tail bk , whi
h term-wise majorizes ak . Then ak =
k=n k=n k=1
nP1 1
P nP1 1
P nP1 1
P
ak + ak ak + bk ak + bk < 1.
k=1 k=n k=1 k=n k=1 k=1
1
P
Example Consider series k2 k . The ratio of two su
essive terms aak+1
k
of
k=1
the series is k2+1 2
k . This ratio is less or equal to 3 starting with k = 3. Hen
e this
P1 2k
series is eventually majorities by geometri
one a3 3k , (a3 = 32 ). This proves
k=0
its
onvergen
e. And now by autore
ursion equation one gets its sum.
Problems.
1
P
1. Prove 0=0
k=1
1 k
P
2. Prove 0 =1
k=1
1
P 1
P 1
P
3. Prove ak = a2 k + a2k+1
k=0 k=0 k=0
1
P 1
P 1
P
4. Prove (ak bk ) = ak bk for
onvergent series
k=1 k=1 k=1
1
P 1
5. Evaluate k(k+1)
k=1
6) + = 1
1) + (1 1 1 1 [( 12 1 1 1
6. Prove (1 2 3 4) + (5 3) + (4 5) + : : :
15
1 2k
P
7. Prove the
onvergen
e of k!
k=0
1 1000k
P
8. Prove the
onvergen
e of k!
k=1
1 k1000
P
9. Prove the
onvergen
e of 2k
k=1
10. Prove that qn < n(11 q) for 0 < q < 1
11. Prove that for any positive q < 1 there is an n that qn < 1
2
1 1
P
12. Prove k! 2
k=1
1
P 1
13. Evaluate k(k+2)
k=1
1 1
P
14. Prove the
onvergen
e of the Euler series k2
k=1
P 1
1 P 1 P
P 1
15. Prove that aij = aij for aij 0
i=1 j =1 j =1 i=1
16
1.3 Unordered Sums
On the
ontents of the le
ture Our summation theory
ulminates in the
Sum Partition Theorem. This le
ture will
ontribute towards evaluation of the
Euler series in two ways: we prove its
onvergen
e, and even estimate its sum
by 2. On the other hand, we will realize that evaluation of the Euler series with
Euler's a
ura
y (10 18 ) seems to be beyond human being's strength.
Consider a family fai gi2I of nonnegative numbers indexed by elements of
an arbitrary set I . An important spe
ial
ase of I is the set of pairs of natural
numbers N N . Families indexed by N N are
alled double series. They arise
when one multiply one series
P by another one.
Any sum of the type ai , where K is a nite subset of I is
alled a subsum
i2K
of fai gi2I over K .
Denition. The least number majorizing all subsums
P of fai gi2I over nite
subsets is
alled its (ultimate) sum and denoted by ai
i2I
One-for-All and All-for-One prin
iples for non-ordered sum are obtained
from
orresponding prin
iples for ordered sum by repla
ing "partial sums" by
"nite subsums".
17
In parti
ular, for K = I , Lemma 2 turns into
P P
Lemma 3. ai = ai [i 2 I ℄
i2I i 2I
P F
Lemma 4. [i 2 Ik ℄ = [i 2 IK ℄ for all i 2 I i IK = k2K Ik
k2K
18
1 1
P
Estimation of the Euler series Let us
ompare the Euler series with 2k ,
k=0
1
P 2n+1
P 1
blo
ked by f2n g to ak . The sum 1 n
k2
onsists of 2 summands, all of
k=1 k=2n
whi
h are less then the rst one, whi
h is 1 n 1 1
22n . As 2 22n = 2n , it follows that
P1
an 21n for ea
h n. Summing these inequalities, one gets ak 2.
k=1
Now let us estimate how many terms of Euler's series one needs to take
into a
ount to nd its sum up to the eighteenth digit. To do this, we need to
1 1
P
k2 2n 1 . To obtain a lower
estimate its tail. The arguments above give 1
k=2n
2n+1
P 1
estimation, let us remark that all terms of sum 1
ex
eed 22(n1+1) . As the
k2
k=2n
1 1
P
number of summands is 2n , one gets an 412n . Hen
e k2 2n+1 . Sin
e
1
k=2n
210 = 1024 ' 103 , one gets 260 ' 1018 . So, to provide a
ura
y 10 18 one needs
to sum approximately 1018 terms. This task is ina
essible even for a modern
omputer. How did Euler manage to do this? He invented a summation formula
(Euler-Ma
Laurin formula) and transformed this slowly
onvergent series into
non-positive divergent (!) one, whose partial sum
ontaining as little as ten
terms gave 18-digit a
ura
y. The whole
al
ulation took him an evening. To
introdu
e this formula, one needs to know integrals and derivatives. We will do
this later.
Problems.
1
P 1
1
P 1
1 1
P
1. Find (2k)2 and (2k 1)2 , assuming k2 = 2 =6.
k=1 k=1 k=1
1
P
2. Prove
onvergen
e of p1 .
k=1 k k
1 1
P
3. Estimate how many terms of series n3 is ne
essary to sum for
al
u-
k=1
lation of its sum with pre
ision 10 3.
1 1 1
P
4. Estimate the value of 2n k .
k=1
1
P 1
P P
5. Prove equality ak bk = aj bk .
k=0 k=0 j;k2N
6. Estimate how many terms of Harmoni
series gives the sum surpassing
1000
7. Prove the Diri
hlet formula
nX1 kX
+1 n nX1
X
aki = aki :
k=1 i=1 i=1 k=i
19
P 1
8. Evaluate .
i;j 2N 2i 3j
P i+j
9. Evaluate .
i;j 2N 2i 3j
P
10. Represent an unordered sum aij as a double sum.
i+j<n
P ij
11. Evaluate .
i;j 2N i 3j
2
1 P
P 2i
12. Change the summation order in aij .
i=0 j =0
13. Dene by Iverson notation the following fun
tion:
(integral part) [x℄
jxj (module)
sgn x (signum)
n! (fa
torial)
14. Dene only by formulas the expression [p is prime℄
20
1.4 Innite Produ
ts
On the
ontents of the le
ture In this le
ture we be
ome a
quainted with
innite produ
ts. The famous Euler Identity will be proved. We will nd out
that (2) is another name for the Euler series. And we will see how the Euler's
de
omposition of the sine into produ
t works to sum up the Euler Series.
Denition. Produ
t of an innite sequen
e of numbers fak g, su
h that
Q
ak 1
for all k is dened as the least number majorizing all partial produ
ts nk=1 ak =
a1 a2 : : : a n
A sequen
e of natural numbers is
alled essentially nite if all but nitely
many of its elements are equal to zero. Denote by N 1 the set of all essentially
nite sequen
es of natural numbers.
P1
Theorem 1. For any given sequen
e of positive series ajk , j = 1; 2; : : : , su
h
k=0
that aj0 = 1 for all j one has
1X
Y 1 X 1
Y
(1.4.1) ajk = ajkj
j =1 k=0 fkj g2N1 j=1
Summands on the right-hand side of (1.4.1) usually
ontain fa
tors whi
h
are less than one. But ea
h of the summands
ontains only nitely many fa
tors
dierent from 1. So the summands are in fa
t nite produ
ts.
Proof. For a sequen
e fkj g 2 N 1 dene its length as maximal j for whi
h kj 6= 0
and its maximum as the value of its maximal term. The length of zero sequen
e
is dened as 0. P Q1
Consider a nite subset S N 1 . Consider the partial sum j
k=1 akj .
fkj g2S
To estimate it, denote by L the maximal length of elements of S and denote by
M the greatest of maxima of fkj g 2 S . In this
ase
X Y 1 X Y L X Y L Y L X M Y1X 1
ajkj = ajkj ajkj = ajk ajk ;
fkj g2S j=1 fkj g2S j=1 fkj g2NLM j=1 j =1 k=0 j =1 k=0
where N LM denotes the set of all nite sequen
es fk1 ; k2 ; : : : kLg of natural num-
bers su
h that ki M . By All-for-One this implies one of the required inequal-
ities, namely, .
To prove the opposite inequality, we prove that for any natural L one has
Y L X 1 X Y L
(1.4.2) ajk = ajkj ;
j =1 k=0 fkj g2NL j=1
where N L denotes the set of all nite sequen
es fk1 ; : : : kL g of natural numbers.
The proof is by indu
tion on L.
Lemma 2. For any families fai gi2I fbj gj2J of nonnegative numbers, one has
X X X
(1.4.3) ai bj = ai bj
i2I j 2J (i;j )2I J
21
F
Proof. Sin
e I J = fig J by the Sum Partition Theorem one gets:
i2I
(1.4.4)
X X X XX X X X X
ai bj = ai bj = ai bj = ai bj = bj ai
(i;j )2I J i2I (i;j )2figJ i 2I j 2J i 2I j 2J j 2J i2I
22
2n+1
P 1 2n+1
P 1
k
as 1 1 n 1 1
2n+1 = 2 2n+1 = 2 . That is why the harmoni
series
k=2n k=2n
diverges. A Diri
hlet series for s < 1 term-wise majorizes the Harmoni
series
and so diverges.
1 1
P
Riemann -fun
tion Fun
tion (s) = ns is
alled Riemann -fun
tion.
k=1
It is of great importan
e in the number theory.
The simplest appli
ation of the Euler's Identity represents Euler's proof of
P1 1
innity of the set of primes. The divergen
e of the harmoni
series k implies
k=1
the Euler Produ
t has to
ontain innitely many fa
tors to diverge.
EulerP proved essentially more exa
t result: the series of re
ipro
al primes
diverges p1 = 1.
Summing via multipli
ation Multipli
ation of series gives rise to a new
1
P
approa
h to evaluating their sums. Consider the geometri
series xk . Then
k=0
1 !2 1 1
X X X X X
xk = xj xk = xj xk = (k + 1)xm
k=0 j;k2N2 m=0 j +k=m m=0
1
P 1
P
As xk = 1
1 x one gets (k + 1)xk = 1
(1 x)2 .
k=0 k=0
x 1
Y x2
1 1X
Y 1 x 2j
(1.4.7) = 1 =
sin x k=1 k2 2 k=1 j =0
k j 2j
2
1
P 2k+1
To avoid negative numbers, we interpret the series ( 1)k (2xk+1)! in the
P1 x4k+1 P 1k=0x4k+3
se
ond of formulas (1.4.6) as the dieren
e (4k+1)! (4k+3)! . Substituting
k =0 k =0
this expression for sin x in sinx x and
an
elling out x, we get
!j
x 1 1
X 1
X x 2k
(1.4.8) = 1 = ( 1)k+1
sin x 1 P 2k
( 1)k+1 (2kx+1)! j =0 k=1 (2k + 1)!
k=1
23
All terms on the right-hand side starting with j = 2 are divisible2 by x4 . Conse-
quently the only summands with x2 on the right-hand side is x6 . On the other
hand in (1.4.7) after expansion into sum by Theorem 1.4.1 terms with x2 give
1 x2
P 1 1 2
P
the series 2
k 2 . Comparing these results, one gets k2 = 6 .
k=1 k=1
Problems.
Q1
1. Prove n=1 1:1 = 1
2. Prove identity 1
Q Q1
n=1 an = ( n=1 an ) (an 1)
2 2
3. Does 1 (1 + 1 )
onverge?
Q
n=1 n
Q1 n2
4. Evaluate n=2 n2 1
1
Q
5. Prove divergen
e (1 + k1 )[k is prime℄
1
Q1 n(n+1)
6. Evaluate n=3 (n 2)(n+3)
Q1 n2 1
7. Evaluate n=2 n2 4
Q1 1
8. Evaluate n=1 (1 + n(n+2) )
Q1 (2n+1)(2n+7)
9. Evaluate n=1 (2n+3)(2n+5) )
Q1 n3 +1
10. Evaluate n=2 n3 1
Q1 P1 1
k=2 (1 + k2 )
11. Prove inequality 1
k2
k=2
Q 4k 2
12. Prove
onvergen
e of the Wallis produ
t 4k 2 1
1 1
P
13. Evaluate k4 applying (1.4.6).
k=1
Q1 2
14. Prove n=2 nn+1 2 <1
15. Multiply geometri
series by itself and get a power series expansion for
(1 x) 2
P1 (n)
16. Dene (n) as the number of divisors of n, then 2 (x) =
k=1 nx
17. Dene (n) as the number of numbers whi
h are less than n are relatively
(x 1) P 1 (n)
prime to n, then =
(x) k=1 nx
18. Dene (n) (Mobius fun
tion) as follows: (1) = 1, (n) = 0, if n is
divisible by the square of a prime number, (n) = ( 1)k , if n is the
1 1 (n)
P
produ
t of k dierent prime numbers. Then =
(x) k=1 nx
24
P1 [k is prime℄
19. Prove k =1
k=1
20. Prove identity 1
Q 2n 1
n=0 (1 + x ) = 1 x
25
1.5 Teles
opi
sums
On the
ontent of this le
ture. In this le
ture we learn the main se
ret of
the elementary summation theory. We will evaluate series via its partial sums.
We introdu
e fa
torial powers, whi
h are easy to sum. Following Stirling we
expand 1+1x2 into a series of negative fa
torial powers and apply this expansion
to evaluate the Euler series with Stirling's a
ura
y 10 8.
1
P 1
The series k(k+1) .
In the rst le
ture we have
al
ulated innite sums
k=1
dire
tly without invoking partial sums. Now we present a dual approa
h to
summing series. A
ording this approa
h at rst one nds a formula for the
n-th partial sum and then substitutes in this formula innity instead of n. The
P1 1
series k(k+1) gives a simple example for this method. A key to sum it up is
k=1
the following identity
1 1 1
= :
k(k + 1) k k + 1
P1 1
Be
ause of this identity k(k+1) turns into the sum of dieren
es
k=1
1 1 1 1 1 1 1
(1.5.1) 1 + + + + +:::
2 2 3 3 4 n n+1
Its nth partial sum is equal to 1 1
n+1 . Substituting in this formula n = +1,
one gets 1 as its ultimate sum.
Teles
opi
sums. The sum (1.5.1) represents a teles
opi
sum . This name is
n
P
used for sums of the form (ak ak+1 ). The value of su
h a teles
opi
sum is
k=0
determined by the values of the rst and the last of ak , similarly to a teles
ope,
whose thi
kness is determined by the radii of the external and internal rings.
Indeed,
n
X n
X n
X n
X nX1
(ak ak+1 ) = ak ak+1 = a0 + ak ak+1 an+1 = a0 an+1 :
k=0 k=0 k=0 k=1 k=0
The same arguments for innite teles
opi
sum give
X1
(1.5.2) (ak ak+1 ) = a0 :
k=0
P1 P1 1
But this proof works only if ak < 1. This is untrue for k(k+1) , owing to
k=0 k=1
divergen
e of the Harmoni
series. But the equality 1.5.2 holds also if ak tends
to 0 as k tends to innity. Indeed, in this
ase a0 is the least number majorizing
P1
all a0 an , the nth partial sums of ak .
k=0
26
Dieren
es. For a given sequen
e fak g one denotes by fak g the sequen
e
of dieren
es ak = ak+1 ak and
alls latter sequen
e as dieren
e of fak g.
The main formula of elementary summation theory is
nX1
(1.5.3) ak = an a0
k=0
1
P
To teles
ope a series ak it is suÆ
ient to nd a sequen
e fAk g su
h that
k=0
nP1
Ak = ak . On the other hand the sequen
e of sums An = ak has dieren
e
k=0
An = an . Therefore, we see that to teles
ope a sum is equivalent to nd a
formula for partial sums. This lead to
on
ept of teles
opi
fun
tion. For a
fun
tion f (x) wePintrodu
e its dieren
e f (x) as f (x + 1) f (x). A fun
tion
f (x) teles
opes ak , if f (k) = ak for all k.
Often the sequen
e fak g that we would like to teles
ope has the form ak =
f (k) for some fun
tion. Then we are sear
hing for a teles
opi
fun
tion F (x)
for f (x), i.e. a fun
tion su
h that F (x) = f (x).
To evaluate the dieren
e of a fun
tion is usually mu
h easier than to tele-
s
ope it. For this reason one has evaluated the dieren
es of all basi
fun
tions
and organized a table of dieren
es. In order to teles
ope a given fun
tion, look
in this table to nd a table fun
tion whose dieren
e
oin
ides with or is
lose
to given fun
tion.
For example, the dieren
es of xn for n 3 are x = 1, x2 = 2x + 1,
P1 2
x3 = 3x2 + 3x + 1. To teles
ope k we
hoose in this table x3 . Then
k=1
x3 2 3 2
x2 = x + 31 = 2x x6 . Therefore, x2 = x3 x2 + x6 . This
3
immediately implies the following formula for sums of squares:
nX1
2n3 3n2 + n
(1.5.4) k2 =
k=1
6
Fa
torial powers The usual powers xn have a
ompli
ated dieren
es. So-
alled fa
torial powers xk have simpler dieren
es. For any number x and any
natural number k, let xk denote x(x 1)(x 2) : : : (x k + 1), and by x k we
1 0
denote (x+1)(x+2) :::(x+k) . At last we dene x = 1. The fa
torial power satises
the following addition law
(1.5.5) xk+m = xk (x k)m
We leave to the reader to
he
k this rule for all integers m, k. The power nn for
a natural n
oin
ides with the fa
torial n! = 1 2 3 n. The main property of
fa
torial powers is given by
(1.5.6) xn = nxn 1
27
Applying this formula one
an easily teles
ope any fa
torial polynomial, i.e.
an expression of the form a0 + a1 x1 + a2 x2 + a3 x3 + : : : an xn . Indeed, the
expli
it formula for the teles
oping fun
tion is a0 x1 + a21 x2 + a32 x3 + a43 x4 +
n xn+1 . Therefore, another strategy to teles
ope xk is to represent it as a
: : : na+1
fa
torial polynomial.
For example, to represent x2 as fa
torial polynomial,
onsider a + bx +
x2 ,
a general fa
torial polynomial of degree 2. We are looking for x2 = a + bx +
x2 .
Substituting x = 0 in this equality one gets a = 0. Substituting x = 1, one
gets 1 = b, and nally for x = 2 one has 4 = 2 + 2
. Hen
e 2
3 = 1. As result
x2 = x + x2 . And the teles
oping fun
tion is given by x2 + x3 = 12 (x2 x) +
1 (x(x2 3x + 2)) = 1 (2x3 3x2 + x). And we have on
e again proved the
3 6
formula (1.5.4).
Stirling Estimation of the Euler series. We will expand (1+1x)2 into a series
of negative fa
torial powers in order to teles
ope it. A natural rst approxima-
tion to (1+1x2 ) is x 2 = (x+1)(1 x+2) . We represent (1+1x)2 as x 2 + R1 (x), where
R1 (x) = (1+1x)2 x 2 = (x+1)12 (x+2) . The remainder R1 (x) is in a natural way
approximated by x 3 . If R1 (x) = x 3 + R2 (x) then R2 (x) = (x+1)2 (x2+2)(x+3) .
Further, R2 (x) = 2x 4 + R3 (x), where R3 (x) = (x+1)2 (x+2)( 2 3 3!x 4
x+3)(x+4) = x+1 .
The above
al
ulations led to
onje
ture
nX1
1 n!x n 1
(1.5.7) = k !x k 2
+ :
(1 + x)2 k=0 x+1
n 1
This
onje
ture is easily proved by indu
tion. The remainder Rn (x) = n!xx+1
nP1
represents the dieren
e (1+1x)2 k!x 2 k . Owing to the inequality x 1 n
k=0
1
(n+1)! ,whi
h is valid for all x 0, the remainder de
reases to 0 as n in
reases
to innity. This implies
Theorem 1. For all x 0 one has
1 X1
(1.5.8) = k !x 2 k
(1 + x)2 k=0
1 1
P
To
al
ulate (1+k)2 , repla
e all summands by the expressions (1.5.7).
k=p
1 nP1
P 1 n
We will get m!k 2 m + n!kk+1 . Changing the order of summa-
k=p m=0
nP1 1
P P1 n!k 1 n x 1 m
tion we have m! k 2 m + k+1 . Sin
e 1+m teles
opes the se-
m=0 k=p k=p
2 m 1 2 m p 1 m
P
quen
e fk g, we have k = 1+m . Denote the sum of remainders
k=p
1 n!k 1 n
P
k+1 by R(n; p). Then for all natural p and n one has
k=p
1
X 1 nX1
p 1 m
(1.5.9) 2 = m ! + R(n; p)
k=p (1 + k ) m=0 1+m
28
For p = 0 and n = +1, the right-hand side turns into the Euler series, and
1 n
one
ould get a false impression that we get nothing new. But k 2 n kk+1
(k 1) 2 n, hen
e
n!p 1 n X 1 X1 n!(p 1) 1 n
= n!k 2 n R(n; p) n!(k 1) 2 n = :
1+n k=p k=p 1+n
Sin
e (p 1) 1 n
p 1 n
= (1 + n)(p 1) 2 n
, there is a 2 (0; 1) su
h that
n!p 1 n 2 n
R(n; p) = + n!(p 1) :
1+n
Finally we get:
1
X 1 X p 1
1 nX1
p 1 k 2 n
(1.5.10) = + k ! + n!(p 1)
k=1
k2 k=0 (1 + k)2 k=0 1 + k
For p = n = 3 this formula turns into
X1 1 1 1 1 1 1
k 2 = 1 + 4 + 9 + 4 + 40 + 180 + 420
k=1
For p = n = 10 one gets R(10; 10) 10!9 12. After
an
ellations one has
10 1
1 . This is approximately 2 10 8 . Therefore P 1
211121314151719 (k+1)2 +
k=0
10P1 k!10 1 k
1+k is inferior than the sum of the Euler series by only 2 10 8. In
k=0
1 1
P
su
h a way one
an in one hour
al
ulate eight digits of k2 after the de
imal
k=1
point. It is not a bad result, but it is still far from Euler's 18 digits. For p = 10
to provide 18 digits one has to sum essentially more than 100 terms of the series.
This is a bit too mu
h for a person, but is possible for a
omputer.
Problems.
P 3
1. Teles
ope k
2. Represent x4 as a fa
torial polynomial.
1 1
P
3. Evaluate k(k+2)
k=1
1
P 1
4. Evaluate k(k+1)(k+2)(k+3)
k=1
5. If ak bk for all k and a1 b1 then ak bk for all k
6. (x + a)n = n(x + a)n 1
3 nP1 3
7. Prove Ar
himedes's inequality n3 k2 (n+1)
3
k=1
29
1 k
P
8. Teles
ope 2k
k=1
1
P
9. Prove the inequalities n1 1
k2 n+1
1
k=n+1
10. Prove that the degree of P (x) is less than the degree of P (x) for any
polynomial P (x).
11. Relying on 2n = 2n , prove that P (n) < 2n eventually for any polynomial
P (x).
1
P
12. Prove k!(x 1) 1 k = x1
k=0
30
1.6 Complex series
On the
ontents of the le
ture. The
omplex numbers hide the key to the
Euler Series. The summation theory developed for positive series now extends
to
omplex series. We will see that
omplex series
an help to sum real series.
Cubi
equation Complex numbers arose in
onne
tion with solution of
ubi
equation. The substitution x = y a redu
es the general
ubi
equation x3 +
3
ax2 + bx +
= 0 to
(1.6.1) y3 + py + q = 0
The redu
ed equation one solves by the following tri
k. One looks for a root in
the form y = + . Then ( + )3 + p( + ) + q = 0 or 3 + 3 + 3 ( +
) + p( + ) + q = 0 The latter equality one redu
es to the system
3 + 3 = q
(1.6.2)
3 = p
Raising the se
ond equation into
ube one gets
3 + 3 = q
(1.6.3)
273 3 = p3
Now 3 , 3 are roots of the quadrati
equation
p3
(1.6.4) x2 + qx
27
alled a resolution of the original
ubi
equation. Sometimes the resolution has
no roots, while the
ubi
equation always has a root. Nevertheless one
an
evaluate a root of the
ubi
equation with the help of its resolution. To do this
one simply ignores that numbers under square roots are negative.
For example
onsider the following
ubi
equation
3 1
(1.6.5) x3 x =0
2 2
Then (1.6.2) turns into
1
3 + 3 =
(1.6.6) 2
1
3 3 =
8
q
Corresponding resolution is t2 2t + 18 = 0 and its roots are t1;2 = 4
1 1 1 =
1 1 p 1. Then the desired root of
ubi
equation is given by
16 8
4 4
r r
1 p 1 p 1
q
p q
p
(1.6.7) 3 (1 + 1) + 3 (1 1) = p 3 1+ 1+ 3 1 1
4 4 34
31
It turns out that the latter expression one uniquely interprets as a real number
whi
h is a root of the equation 1.6.5. To evaluate it
onsider the following
expression
q
3 p q
3 p q p q p
(1.6.8) (1 + 1)2 (1 + 1) 3 (1 1) + 3 (1 1)2
Sin
e
p p p 2 p p
(1 + 1)2 = 12 + 2 1+ 1 =1+2 1 1=2 1;
p p p3 p p p3 p p
thepleft summand of (1.6.8)
p p is equal top 3 2 1 = 2 3 1= 2 1= 3
32 1: Similarly (1 1) 2 = 2 1, and the right summand of (1.6.8)
pp p p p 2
turns into 3 2p 1. Finally (1 + 1)(1 1) = 12 1 = 2 and pthe
entral one is 3 2. As result the whole expression (1.6.8) is evaluated as 3 2.
On the other hand one evaluates the produ
t of (1.6.7) and (1.6.8) by usual
formula as the sum of
ubes
p p p p p
p31 ((1+ 1)+(1 1)) = p 3
1
((1+1)+( 1) 1)) = p
3
1
(2+0) = 3 2:
4 4 4
p
32
Consequently (1.6.7) is equal to p 3 2 = 1. And 1 is a true root of (1.6.5).
p
Arithmeti
of
omplex numbers. In the sequel we use i instead of 1.
There are two basi
ways to represent a
omplex number. Representation z =
a + ib, where a and b are real numbers we
all Cartesian form of z . The numbers
a and b are
alled respe
tively real and imaginary parts of z and denoted by
Re z and by Im z respe
tively. Addition and multipli
ation of
omplex numbers
are dened via their real and imaginary parts as follows
(1.6.9) Re(z1 + z2 ) = Re z1 + Re z2
(1.6.10) Im(z1 + z2 ) = Im z1 + Im z2
(1.6.11) Re(z1 z2 ) = Re z1 Re z2 Im z1 Im z2
(1.6.12) Im(z1 z2 ) = Re z1 Im z2 + Im z1 Re z2
Trigonometri
form of a
omplex number is z = (
os +i sin ), where 0
is
alled module or absolute value of a
omplex number z and is denoted jz j, and
is
alled its argument. The argument of a
omplex number is dened modulo
2. We denote by Arg(z ) the set of all arguments of z , and by arg z elements of
Arg z whi
h satisfy the inequalities < arg z . So arg z is uniquely dened
for all
omplex numbers. The arg z is
alled prin
ipal argument of z .
The number a bi is
alled
onjugated to z = a + bi and denoted z. One
has zz = jz j2 . This allows us to express z 1 as jzzj2 .
p
If z = a + ib then jz j = a2 + b2 and arg z = ar
tg ab . One represents a
omplex numbers z = a + bi as a points Z of the plane with
oordinates (a; b).
Then jz j is equal to the distan
e from Z to the origin O. And arg z represents
!
the angle between the axis of abs
ises and the ray OZ . Addition of the
omplex
numbers
orresponds to usual ve
tor addition. And the usual triangle inequality
turns into the module inequality :
jz + j jz j + j j:
32
Im z Z
arg z
O Re z
The multipli
ation formula for
omplex numbers in the trigonometri
form is
espe
ially simple:
(1.6.13) r(
os + i sin )r0 (
os + i sin ) = rr0 (
os( + ) + i sin( + ))
Indeed, the left-hand side and the right-hand side of (1.6.13) transform to
rr0 (
os
os sin sin ) + irr0 (sin
os + sin
os ):
That is the module of the produ
t is equal to the produ
t of modules and the
argument of produ
t is equal to the sum of arguments
Arg z1 z2 = Arg z1 Arg z2
Any
omplex number is uniquely dened by its module and argument.
The multipli
ation formula allows us to prove by indu
tion the following:
(Moivre Formula) (
os + i sin )n = (
os n + i sin n)
Sum of
omplex series. Now is the time to extend our summation theory
to series made of
omplex numbers. We extend the whole theory without any
1
P
losses to so
alled absolutely
onvergent series. Series zk with arbitrary
k=1
P1
omplex terms is
alled absolutely
onvergent, if the series jzk j of absolute
k=1
values
onverges.
For any real number x one denes two nonnegative number its positive x+
and negative x parts as x+ = x[x 0℄ and x = x[x < 0℄. The following
identities
hara
terize the positive and negative partes of x
(1.6.14) x+ + x = jxj x+ x =x
Now the sum of an absolute
onvergent series of real numbers is dened as
follows:
1
X X1 X1
(1.6.15) ak = a+k ak
k=1 k=1 k=1
33
That is from the sum all positive summands one subtra
ts the sum of modules
of all negative summands. The series on the right-hand side
onverge, be
ause
P1
a+k jak j, ak jak j and j ak j < 1 .
k=1
P1
For an absolutely
onvergent
omplex series zk we dene the real and
k=1
imaginary parts of its sum separately by formulas
X1 X1 1
X X1
(1.6.16) Re zk = Re zk Im zk = Im zk
k=1 k=1 k=1 k=1
The series in the right-hand sides of these formulas are absolutely
onvergent,
sin
e j Re zk j jzk j and j Im zk j jzk j.
P1 P1
Theorem 1. For any pair of absolutely
onvergent series ak and bk its
k=1 k=1
P1
term-wise sum (ak + bk ) absolutely
onverges and
k=1
1
X 1
X 1
X
(1.6.17) (ak + bk ) = ak + bk
k=1 k=1 k=1
Proof. First, remark that absolute
onvergen
e of the series on the left-hand
side follows from the Module Inequality jak + bk j jak j + jbk j and absolute
onvergen
e of the series on the right-hand side.
Now
onsider the
ase of real numbers. Representing all sums in (1) as
dieren
es of its positive and negative parts and separating positive and negative
terms in dierent sides one transforms (1) into
X1 X1 X1 X1 1
X 1
X
a+k + b+k + (ak + bk ) = ak + bk + (ak + bk )+
k=1 k=1 k=1 k=1 k=1 k=1
But this equality is true due to term-wise addition for positive series and the
following identity,
x + y + (x + y)+ = x+ + y+ + (x + y)
Moving terms around turns this identity into
(x + y)+ (x + y) = (x+ x ) + (y+ y );
whi
h is true due to the identity x+ +x = x.
In the
omplex
ase the equality (1.6.17) splits into two equalities, one for
real and another for imaginary parts. As for real series the term-wise addition
is already proved, we
an write the following
hain of equalities
X1 X1 ! X1 1
X X1 X 1
Re ak + bk = Re ak + Re bk = Re ak + Re bk =
k=1 k=1 k=1 k=1 k=1 k=1
1
X 1
X 1
X
(Re ak + Re bk ) = Re(ak + bk ) = Re (ak + bk );
k=1 k=1 k=1
whi
h proves the equality of real parts in (1.6.17). The same proof works for
imaginary parts.
34
Sum Partition Theorem. An unordered sum of a family of
omplex numbers
is dened by the same formulas (1.6.15) and (1.6.16). Sin
e for positive series
non-ordered sum
oin
ides with the ordered one, we get the same
oin
iden
e
for all absolutely
onvergent series. Hen
e the
ommutativity law holds for all
absolutely
onvergen
e series.
1
P P P
Theorem 2. If I = tj2J Ij and jak j < 1 then j2J i2Ij ai < 1 and
k=1
XX X
(1.6.18) ai = ai
j 2J i2Ij i 2I
P
Proof. At rst
onsider the
ase of real summands. By denition ai =
P + P i2I
a i ai By Sum Partition Theorem positive series one transforms the
i2I i2I P P P P
original sum into a+i ai Now by the Term-Wise Addition ap-
j 2J i2Ij j 2J i2Ij
plied at rst to external and after to internal sums one gets
0 1
X X X XX XX
a+i ai A = (a+i ai ) = ai :
j 2J i2Ij i2Ij j 2J i2Ij j 2J i2Ij
So the Sum Partition Theorem is proved for all absolutely
onvergent real se-
ries. And it immediately extends to absolutely
onvergent
omplex series by its
splitting into real and imaginary parts.
1
P 1
P
Theorem 3 (Term-Wise Multipli
ation). If jzk j < 1 then j
zk j <
k=1 k=1
P1 1
P
1 for any (
omplex)
and
zk =
zk
k=1 k=1
Proof. Term-wise Multipli
ation for positive numbers gives the rst statement
P1 1 1
of the theorem j
zk j = P j
jjzk j = j
j P jzk j. The further proof is divided
k=1 k=1 k=1
into ve
ases.
At rst suppose
is positive and zk real. Then
zk+ =
zk+ and by virtue of
Term-Wise Multipli
ation for positive series we get
1
X 1
X 1
X 1
X 1
X
zk =
zk+
zk =
zk+
zk =
k=1 k=1 k=1 k=1 k=1
!
1
X 1
X 1
X
zk+ zk =
zk
k=1 k=1 k=1
The se
ond
ase. Let
= 1 and zk be real. In this
ase
X1 X1 X1 X1 X1 1
X
zk = ( zk )+ ( zk ) = zk zk+ = zk
k=1 k=1 k=1 k=1 k=1 k=1
The third
ase. Let
be real and zk
omplex. In this
ase Re
zk =
Re zk
and the two
ases above imply the Term-Wise Multipli
ation for any real
.
35
Hen
e
X1 1
X 1
X 1
X 1
X 1
X
Re
zk = Re
zk =
Re zk =
Re zk =
Re zk = Re
zk
k=1 k=1 k=1 k=1 k=1 k=1
The same is true for imaginary parts.
The fourth
ase. Let
= i and zk be
omplex. Then Re izk = Im zk and
Im izk = Re zk . So one gets for real parts
X1 X1 1
X
Re izk = Re(izk ) = Im zk =
k=1 k=1 k=1
1
X 1
X 1
X
Im zk = Im zk = Re i zk
k=1 k=1 k=1
The general
ase. Let
= a + bi with real a, b. Then
1
X X1 X1 X1 X1 X1 1
X
zk = a zk + ib zk = azk + ibzk = (azk + ibzk ) =
zk
k=1 k=1 k=1 k=1 k=1 k=1 k=1
1
P 1
P
Multipli
ation of Series For given two series ak and bk , one denes
k=0 k=0
1
P Pn
their
onvolution as a series
n , where
k = ak bn k .
n=0 k=0
1
P 1
P
Theorem 4 (Cau
hy). For any absolutely
onvergent series ak and bk
k=0 k=0
1
P
the
onvolution
k absolutely
onverges and
k=0
1
X 1
X 1
X
(1.6.19)
k = ak bk
k=0 k=0 k=0
P
Proof. Consider the double series ai bj . Then by Sum Partition Theorem its
i;j
sum is equal to
! ! !0 1
X1 X 1 1
X 1
X 1
X X1
ai bj = bj ai = ai bj A :
j =0 i=0 j =0 i=0 i=0 j =0
P 1 n+1
P P1
On the other hand, ai bj = ak bn k . But the last sum is just the
i;j n=0 k=0
onvolution.
This proof goes through for positive series. In general
ase we have to prove
absolute
onvergen
e of the double series. But this follows from
! !
X1 1
X 1
X
jak j jbk j = j
k j
k=0 k=0 k=0
36
Module Inequality.
X1 1
X
(1.6.20)
zk
j zk j
k=1 k=1
Let zk = xk + iyk . The summation of inequalities jxk j xk jxk j
P1 1 1 P 1 1
gives jxk j xk jxk j, whi
h means xk P jxk j. The
P P
k=1 k=1 k=1 k=1 k=1
0 0
same inequality
for yk . Consider zk = jxk j + ijyk j Then jzk j = jzk j
is true
P 1 P 1 0
and zk zk . Therefore it is suÆ
ient to prove inequality (1.6.20)
k=1 k=1
for zk0 , that is for numbers with non-negative real and imaginary parts. Now
supposing xk ; yk to be nonnegative one gets the following
hain of equivalent
transformations of (1.6.20):
!2 !2 !2
1
X 1
X 1
X
xk + yk j zk j
k=1 k=1 k=1
v
1 u 1 !2 1 !2
X u X X
xk t jzk j yk
k=1 k=1 k=1
v
n u 1 !2 1 !2
X u X X
xk t j zk j yk 8n = 1; 2; : : :
k=1 k=1 k=1
v
u 1 !2 !2
1
X u X n
X
yk t jzk j Re xk 8n = 1; 2; : : :
k=1 k=1 k=1
v
u 1 !2 !2
m
X u X n
X
yk t j zk j xk 8n; m = 1; 2; : : :
k=1 k=1 k=1
n !2 m !2 1 !2
X X X
xk + yk j zk j 8m; n = 1; 2; : : :
k=1 k=1 k=1
v
u N !2 !2
XN u X N
X 1
X
zk
= t
xk + yk jzk j 8N = 1; 2; : : :
k=1 k=1 k=1 k=1
PN N
P 1
P
The inequalities of the last system hold be
ause zk jzk j j zk j .
k=1 k=1 k=1
37
In parti
ular, for z = q(sin + i
os ) the real part of the left-hand side of
nP1
(1.6.21) owing to Moivre Formula turns into qk sin k and the right-hand
k=0
nP1
side turns into qk
os k.So the formula for geometri
progression splits
k=0
into two formulas whi
h allow us to teles
ope some trigonometri
series.
Espe
ially interesting is the
ase with the ratio "n =
os 2n + i sin 2n . In this
ase the geometri
progression
y
li
ally takes the same values, be
ause "nn = 1.
The terms of this sequen
e are
alled the roots of unity , be
ause they satisfy
the equation z n 1 = 0.
Lemma 5.
n
Y
(1.6.22) (z n 1) = (z "kn )
k=1
Proof. Denote by P (z ) the right-hand side produ
t. This polynomial has degree
n has the major
oeÆ
ient 1 and has all "kn as its roots. Then the dieren
e
(z n 1) P (z ) is a polynomial of degree < n whi
h has n dierent roots. Su
h
polynomial has to be 0 by virtue of the following general theorem.
Theorem 6. Number of roots of any nonzero
omplex polynomial does not
ex
eed its degree.
Proof. Proof is by indu
tion on degree of P (z ). Polynomial of degree 1 has form
az + b and the only root ab . Suppose our theorem is proved for any polynomial
of degree < n. Consider a polynomial P (z ) = a0 + a1 z + + an z of degree
n, where
oeÆ
ient are
omplex numbers. Suppose Q
it has at least n roots
z1 : : : zn . Consider the polynomial P (z ) = an nk=1 (z zk ). The dieren
e
P (z ) P (z ) has degree < n and has at least n roots (all zk ). By the indu
tion
hypothesis this dieren
e is zero. Hen
e, P (z ) = P (z ). But P (z ) has only n
roots. Indeed, for any z dierent from all zk one has jz zk j > 0. Therefore
jP (z )j = jan j Qnk=1 jz zk j > 0.
Blo
king
onjugated roots one gets a pure real formula.
(nY
1)=2
2k
(1.6.23) z n 1 = (z 1) z 2 2z
os +1
k=1 n
38
osine series. In parti
ular, for q = 1, one has 1 1 z = 1+
os 1+i sin . To evaluate
the real and imaginary parts one multiplies both numerator and denominator
by 1 +
os i sin . Then one gets (1
os )2 + sin2 = 1 2
os2 +
os2 +
sin2 = 2 2
os as the denominator. Hen
e 1 1 z = 1
os +i sin 1 1
2 2
os = 2 + 2
ot 2 .
And we get two remarkable formulas for the sum of the divergent series
X1 1 X1 1
(1.6.24)
os k = sin k =
ot
k=0 2 k=1 2 2
P1
For = 0 the left series turns into ( 1)k . The evaluation of the Euler series
k=0
via this
osine series is remarkably short, it takes one line. But one has to know
integrals and a something else to justify this evaluation.
Problems.
3
1 1 i i5 + 2 (1 + i)5
1. Find real and imaginary parts for , , , .
1 i 1+i i19 + 1 (1 i)3
p
2. Find trigonometri
form for 1, 1 + i , 3 + i.
3. Prove that z1 z2 = 0 implies either z1 = 0 or z2 = 0.
4. Prove the distributivity law for
omplex numbers.
5. Analyti
ally prove the inequality jz1 + z2 j jz1 j + jz2 j.
nP1
6. Evaluate 1 where zk = 1 + kz .
zk (zk +1) ,
k=1
nP1
7. Evaluate zk2 , where zk = 1 + kz .
k=1
nP1 sin k
8. Evaluate .
k=1 2k
9. Solve z 2 = i.
10. Solve z 2 = 3 4i.
P1 sin 2n
11. Teles
ope 3n .
k=1
12. Prove that the
onjugated to a root of polynomial with real
oeÆ
ient is
the root of the polynomial.
13. Prove that z1 + z2 = z1 + z2 .
14. Prove that z1 z2 = z1 z2 .
15. Solve 8x3 6x 1 = 0
1 sin n
P
16. Evaluate
k=1 2n
39
1 sin 2n
P
17. Evaluate 3n
k=1
1 zk
P
18. Prove absolute
onvergen
e of k! for any z
k=0
1 zk
P
19. For whi
h z the series k absolutely
onverges?
k=1
20. Multiply a geometri
series onto itself several times applying Cau
hy for-
mula
p
21. Find series for 1 + x by method of indenite
oeÆ
ients
1 sin k
P
22. Does series k absolutely
onverge?
k=1
1 sin k
P
23. Does series k2 absolutely
onverge?
k=1
40
Chapter 2
Integrals
41
2.1 Natural Logarithm.
On the
ontents of the le
ture.
In the beginning of Cal
ulus was the Word, and the Word was with
Arithmeti
, and the Word was Logarithm 1
Logarithmi
tables Multipli
ation is mu
h more diÆ
ult than addition. The
Logarithm redu
es multipli
ation to addition. Inventions of logarithms were one
of the great a
hievements of our
ivilization.
In early times, when logarithms was unknown instead of them one used
trigonometri
fun
tions. The following identity
2
os x
os y =
os(x + y) +
os(x y)
an be applied to
al
ulate produ
ts via tables of
osines. To multiply numbers
x and y, one represents them as
osines x =
os a , y =
os b using the
osine
table. Then one evaluates (a + b) and (a b) and nd their
osines in the
table. Finally, the results are summed and divided by 2. That is all. A single
multipli
ation requires 4 sear
hes in the table of
osines, two additions, one
subtra
tion and one division by 2.
A logarithmi
fun
tion l(x) is a fun
tion su
h that l(xy) = l(x)+ l(y) for any
x and y. If one has a logarithmi
table, to evaluate the produ
t xy one has to
nd in logarithmi
table l(x) and l(y) then sum them and nd the antilogarithm
of the sum. This is mu
h easier.
The idea of logarithms arose in 1544, when M. Stiefel
ompared a geometri
and arithmeti
progressions. The addition of exponents
orresponds to the
multipli
ation of powers. Hen
e
onsider a number
lose to 1, say, 1:000001.
Cal
ulate the sequen
e of its powers and pla
e them to the left
olumn. Pla
e
to the right
olumn the
orresponding values of exponents, whi
h are just the
line numbers. The logarithmi
table is ready.
Now to multiply two numbers x and y, one nds them (or their approxima-
tions) in the left
olumn of the logarithmi
table, reads their logarithms from
the right
olumn. Sum the logarithms and nd the value of the sum in the right
olumn. Next to this sum in the left
olumn the produ
t xy stands. The rst
tables of su
h logarithms were
omposed by John Napier in 1614.
42
a b
following sums:
nX1 nX1
(2.1.1) f (xk )jxk j f (xk+1 )jxk j (where xk = xk+1 xk )
k=0 k=0
We will
all the rst of them the re
eding sum and the se
ond, the advan
ing
sum of the sequen
e fxk g for the fun
tion f . If the fun
tion f is monotone
the area of the
urvilinear trapezium is
ontained between these two sums.
To see this,
onsider the following step-gures: [kn=01 [xk ; xk+1 ℄ [0; f (xk )℄ and
[kn=01 [xk ; xk+1 ℄ [0; f (xk+1 )℄. If f and fxk g both in
rease or both de
rease the
rst step-gure is
ontained in the
urvilinear trapezium and the se
ond step-
gure
ontains the trapezium with possible ex
eption of a verti
al segments
[a [0; f (a)℄ or [b [0; f (b)℄. If one of f and fxk g in
reases and the other
de
reases, then the step-gures swit
h the roles. The ba
k sum equals the area
of the rst step-gure, and the advan
ing sum equals the area of the se
ond
one. Thus we have proved the following lemma.
Lemma 1. Let f be a monotone fun
tion, S the area of the
urvilinear trapez-
ium under the graph of f over [a; b℄. Then for any sequen
e fxk gnk=0 exhausting
nP1 nP1
[a; b℄ the area S is
ontained between f (xk )jxk j and f (xk+1 )jxk j.
k=0 k=0
43
For a natural a, one has 1 1qaq+1 = 1+q+q12 +:::qa . As q tends to 1 both sums
a+1
onverge to Ba+1 . This is the area of the
urvilinear trapezium. Let us remark
that for a < 0 this trapezium is unbounded, nevertheless it has nite area if
a > 1.
If a < 1, then
onsider an interval in the form [B; 1℄. Choose a positive q >
1. Then the innite geometri
progression B; Bq; Bq2 ; Bq3 : : : exhausts [B; 1℄
and the values of fun
tion for this sequen
e also form a geometri
progression
B a ; qa B a ; q2a B a ; q3a B a ; : : : . The re
eding and advan
ing sums are
X1 X1 B a+1 (q 1)
B a qka (qk+1 B qk B ) = B a+1 (q 1) qk(a+1) =
k=0 k=0 1 qa+1
X1 X1 B a+1 (q 1)qa
B a q(k+1)a (qk+1 B qk B ) = B a+1 (1 q) q(k+1)(a+1) = :
k=0 k=0 1 qa+1
44
1 x
passing through the end points of the interval, is
alled a hyperboli
trapezium
over [a; b℄.
The area of hyperboli
trapezium over [1; x℄ with x > 1 is
alled the natural
logarithm of a x, it is denoted by ln x. For a positive number x < 1 its logarithm
is dened as the negative number whose absolute value
oin
ides with the area
of hyperboli
trapezium over [x; 1℄. At last, ln 1 is dened as 0.
Theorem 3 (on logarithm). Natural logarithm is an in
reasing fun
tion de-
ned for all positive numbers. For ea
h pair of positive numbers x, y
(2.1.2) ln xy = ln x + ln y:
Proof. Consider the
ase x; y > 1. The dieren
e ln xy ln y is the area of the
hyperboli
trapezium over [y; xy℄. And we have to prove that it is equal to ln x,
the area of trapezium over [1; x℄. Choose a large number n. Let q = x1=n . Then
qn = x. The nite geometri
progression fqk gnk=0 exhausts [1; x℄ . Then the
re
eding and advan
ing sums are
nX1 nX1
n(q 1)
(2.1.3) q k (qk+1 qk ) = n(q 1) q k 1 (qk+1 qk ) = :
k=0 k=0
q
Now
onsider the sequen
e fxqk gnk=0 exhausting [x; xy℄. Its re
eding sum
nX1
x 1 q k (xqk+1 xqk ) = n(q 1)
k=0
just
oin
ides with the re
eding sum (2.1.3) for ln x. The same is true for the
advan
ing sum. As a result we obtain for any natural n one has the following
inequalities:
n(q 1) n(q 1)
(2.1.4) n(q 1) ln x n(q 1) ln xy ln y
q q
This implies that j ln xy ln x ln yj does not ex
eed the dieren
e between
the the re
eding and advan
ing sums. The statement of Theorem 3 in the
ase
x; y > 1 will be proved when we will prove that this dieren
e
an be made
arbitrarily small by a
hoi
e of n. This will be dedu
ed from the following
general lemma.
45
Lemma 4. Let f be a monotone fun
tion over interval [a; b℄ and a sequen
e
fxk gnk=0 that exhausts [a; b℄. Then
nX1 nX1
(2.1.5)
f (xk )xk f (xk+1 )xk
jf (b) f (a)j max
k<n
jxk j
k=0 k=0
The proof of the lemma is a straightforward
al
ulation. To shorten the
notation, set f (xk ) = f (xk+1 ) f (xk ).
nX1 nX1 nX1
f (xk )xk f (xk+1 )xk =
f (xk )xk
k=0 k=0 k=0
nX1 nX1
jf (xk )j max jxk j = max jxk j jf (xk )j =
k=0 k=0
n 1
X
max j j
xk f (xk ) = max jxk jjf (b) f (a)j
k=0
nP1 nP1
The equality f (xk ) = jf (xk )j holds, as f (xk ) have the same
k=0 k=0
signs due to the monotoni
ity of f .
The value max jxk j is
alled maximal step of the sequen
e fxk g. For the
sequen
e fqk g of [1; x℄ its maximal step is equal to qn qn 1 = qn (1 q 1 ) =
x(1 q)=q. It tends to 0 as q tends to 1. In our
ase jf (b) f (a)j = 1 x1 < 1.
By Lemma 4 the dieren
e between the re
eding and advan
ing sums
ould be
made arbitrarily small. This
ompletes the proof in the
ase x; y > 1.
Consider the
ase xy = 1, x > 1. We need to prove the following
(inversion rule) ln 1=x = ln x
As above, put qn = x > 1. The sequen
e fq k gnk=0 exhausts [1=x; 1℄. The
nP1 nP1
orresponding re
eding sum qk+1 (q k q k 1 ) = (q 1) = n(q 1)
k=0 k=0
oin
ides with its
ounterpart for ln x. The same is true for the advan
ing one.
The same arguments as above prove j ln 1=xj = ln x. The sign of ln 1=x is dened
as minus be
ause 1=x < 1. This proves the inversion rule.
Now
onsider
ase x < 1, y < 1. Then 1=x > 1 and 1=y > 1 and by the rst
ase ln 1=xy = (ln 1=x+ln 1=y). Repla
ing all terms of this equation a
ording to
the inversion rule, one gets ln xy = ln x ln y and nally ln xy = ln x + ln y.
The next
ase is x > 1, y < 1, xy < 1. Sin
e both 1=x and xy are less then
1, then by the previous
ase ln xy + ln 1=x = log xy
x = ln y . Repla
ing ln 1=x by
ln x one gets ln xy ln x = ln y and nally ln xy = ln x + ln y.
The last
ase, x > 1, y < 1, xy > 1 is proved by ln xy + ln 1=y = ln x and
repla
ing ln 1=y by ln y.
Base of logarithm Natural or hyperboli
logarithms are not the only log-
arithmi
fun
tion. Other popular logarithms are de
imal ones. In
omputer
s
ien
e one prefers binary logarithms. Dierent logarithmi
fun
tions are dis-
tinguished by its bases . The base of a logarithmi
fun
tion l(x) is dened as a
number b for whi
h l(b) = 1. Logarithms with the base b are denoted by logb x.
46
What is the base of natural logarithms? This is the se
ond most important
on-
stant in mathemati
s (after ). It is an irrational number denoted by e whi
h is
equal to 2:71828182845905 : : : . It was Euler who introdu
ed these number and
notation.
Well, e is the number su
h that the area of hyperboli
trapezium over [1; e℄
is 1. Consider geometri
progression qn for q = 1 + n1 . All summands in
the
orresponding hyperboli
re
eding sum for this progression are equal to
qk+1 qk = q 1 = 1 . Hen
e the re
eding sum for the interval [1; q n ℄ is equal
qk n
to 1 and it is greater than log qn . Consequently e > qn . The summands of
k+1 k
the advan
ing sum in this
ase are equal to q qk+1q = 1 1q = n+1 1 . Hen
e
the advan
ing sum for the interval [1; qn+1 ℄ is equal to 1. It is less than the
orresponding logarithm. Consequently, e < qn+1 . Thus we have proved the
following estimates for e:
1 n 1 n+1
(2.1.6) 1+ <e< 1+
n n
We see that 1 + n1 n rapidly tends to e as n tends to innity.
Problems.
1. Prove that ln x=y = ln x ln y
2. Prove that ln 2 < 1
3. Prove that ln 3 > 1
4. Prove that x > y implies ln x > ln y.
5. Is ln x bounded?
1 1
6. Prove that < ln(1 + 1=n) <
n+1 n
x
7. Prove that < ln(1 + x) < x
1+x
8. Prove Fermat theorem for a = 1=2; 1=3; 2=3
9. Prove unboundedness of lnnn .
n+1
10. Compare 1 + n1 n and 1 + n+1
1
48
2.2 Denite Integral.
On the
ontents of the le
ture. Areas of
urvilinear trapezia play an ex-
traordinary important role in mathemati
s. They generate a key
on
ept of
al
ulus | the
on
ept of integral.
Rb
Three Basi
Rules For a nonnegative fun
tion f its integral f (x) dx along
a
the interval [a; b℄ is dened just as the area of the
urvilinear trapezium below
the graph of f over [a; b℄. We allow a fun
tion to take innite values. Let us
remark that
hanging of the value of fun
tion in one point does not ae
t the
integral, be
ause the area of the line is zero. That is why we allow the fun
tions
under
onsideration to be undened in a nite number of points of the interval.
Immediately from the denition one gets the following three basi
rules of
integration :
Rb
Rule of
onstant f (x) dx =
(b a), if f (x) =
for x 2 (a; b)
a
Rb Rb
Rule of inequality f (x) dx g(x) dx, if f (x) g(x) for x 2 (a; b)
a a
R
Rb R
Rule of partition f (x) dx = f (x) dx + f (x) dx for b 2 (a;
)
a a b
Pie
ewise
onstant fun
tions. A fun
tion f (x) is
alled partially
onstant
on a partition fJk gnk=1 of [a; b℄ if it is
onstant on any Jk . The Rules of Constant
and Partition immediately imply:
Zb n
X
(2.2.1) f (x) dx = f (Jk )jJk j:
a k=1
Proof. Indeed, the integral splits into the sum of integrals over Jk = [xk 1 ; xk ℄,
and the fun
tion takes the value f (Jk ) in (xk 1 ; xk ).
A fun
tion is
alled pie
ewise
onstant over an interval, if it is partially
onstant with respe
t to some nite partition of the interval.
Lemma 1. Let f and g be pie
ewise
onstant fun
tions over [a; b℄. Then
Rb Rb Rb
(f (x) g(x)) dx = f (x) dx g(x) dx.
a a a
49
Proof. First, suppose f (x) =
is
onstant on the interval (a; b). Let g takes the
value gk over the interval (xk ; xk+1 ) for an exhausting fxk gnk=0 . Then f (x)+g(x)
Rb nP1
takes values (
+ gk ) over (xk ; xk+1 ). Hen
e (f (x)+ g(x)) dx = (
+ gk )jxk j
a k=0
due to (2.2.1). Splitting this sum and applying (2.2.1) to the both summands,
nP1 nP1 Rb Rb
one gets
jxk j + gk jxk j = f (x) dx + g(x) dx. This proves the
ase
k=0 k=0 a a
of
onstant f .
Now let f be partially
onstant on the partition generated by fxk gnk=0 .
Rb n xRk
P
Then, by the partition rule, (f (x) + g(x)) dx = (f (x) + g(x)) dx. As
a k=1 xk 1
xRk
f is
onstant on any (xk 1 ; xk ), for any k one gets (f (x) + g(x)) dx =
xk 1
xRk xRk
f (x) dx + g(x) dx. Summing up these equalities one
ompletes the
xk 1 xk 1
proof of Lemma 1 for the sum.
The statement about dieren
es follows from the addition formula applied
to g(x) and f (x) g(x).
Lemma 2. For any monotone nonnegative fun
tion f on the interval [a; b℄ and
for any " > 0 there is su
h pie
ewise
onstant fun
tion f" su
h that f" f (x)
f" (x) + ".
1
P
Proof. f" (x) = k"[k" f (x) < (k + 1)"℄.
k=0
50
ex
eed
Zb Zb
(2.2.3) (f " (x) f" (x)) dx + (g" (x) g" (x)) dx 2"(b a):
a a
Hen
e, for any positive "
Zb Zb Zb
(2.2.4) (f (x) + g (x)) dx f (x) dx g(x) dx < 2"(b a):
a a a
This implies that the left-hand side vanishes.
51
n Rb
P
By the the Addition Theorem the left-hand side is equal to fk (x) dx, whi
h
k=1 a
P1 Rb
is a partial sum of fk (x) dx. Then by All-for-One one gets the inequality
k=1 a
1 Rb
P 1
Rb P
fk (x) dx fk (x) dx.
k=1 a a k=1
To prove the opposite inequality for any positive ", we apply Lemma 2
1
P
to nd a pie
ewise
onstant fun
tion F" , su
h that F" (x) fk (x)dx and
k=1
1
Rb P
(fk (x) F" (x)) dx < ". On the other hand, by Lemma 4 one gets
a k=1
1Z b Zb
X
fk (x) dx F" (x) dx:
k=1 a a
P1 Rb 1
Rb P
Together these inequalities imply fk (x) dx + " fk (x) dx. As the
k=1 a a k=1
last inequality holds for all " > 0, it holds also for " = 0
Theorem 6 (Mer
ator,1668). For any x 2 ( 1; 1℄ one has
1
X ( 1)k+1 xk
(2.2.6) log(1 + x) =
k=1 k
Rx k+1
Proof. Consider x 2 [0; 1). Sin
e tk dt = tk+1 due to the Fermat Theorem 2,
0
P1 k
term-wise integration of the geometri
series t over interval [0; x℄ for x < 1
k=0
Rx 1 Rx k
P P1 xk+1
gives 1 1 t dt = t dt = k+1 .
0 k=0 0 k=0
Rx 1
Lemma 7. 1 t dt = ln(1 x)
0
Proof of Lemma. Constru
t a translation of the plane whi
h transforms the
urvilinear trapezium below 1 1 t over [0; x℄ into the trapezium for ln(1 x).
Indeed, the re
e
tion of the plane ((x; y) ! (2 x; y)) along the line x = 1
transforms this trapezium to the
urvilinear trapezium under x 1 1 over [2 x; 2℄.
The parallel translation by 1 to the left of the latter trapezium (x; y) ! (x 1; y)
transform it just in logarithmi
trapezium for ln(1 x).
The Lemma proves the Mer
ator Theorem for negative x. To prove it for
positive x, set fk (x) = x2k 1 x2k . All fun
tions fk are nonnegative on [0; 1℄ and
1
P
fk (x) = 1+1 x . Term-wise integration of this equality over [0; x℄ gives (2.2.6),
k=1
Rx 1 Rx 1
modulo equality 1+t dt =dt. The latter is proved by parallel translation
t
0 1
1 ( 1)k+1 xk
P
of the plane. Let us remark, that in the
ase x = 1 the series k is not
k=1
52
1
P 1
1
P
absolutely
onvergent, and under its sum we mean 2k(2k 1) = ( 2k1 1
k=1 k=1
1
2k ). And the above proof proves just this fa
t.
The arithmeti
mean of Mer
ator's series evaluated at x and x gives Gre-
gory's Series
1 1+x x3 x5 x7
(2.2.7) ln = x + + + + ::::
2 1 x 3 5 7
Gregory's series
onverges mu
h faster than Mer
ator's one. For example,
putting x = 31 in (2.2.7) one gets
2 2 2 2
ln 2 = + + + + ::::
3 33 3 535 7 37
Problems.
Rb Rb
1. Prove that
f (x) dx
a
jf (x)j dx
a
2. Prove the following formulas via pie
e-wise
onstant approximations:
Zb Zb
(multipli
ation formula) f (x) dx = f (x) dx
a a
Zb bZ+
(shift formula) f (x) dx = f (x
) dx
a a+
Za Z0
(re
e
tion formula) f (x) dx = f ( x) dx
0 a
Za Zka
1 x
(
ompression formula) f (x) dx = f dx
k k
0 0
2R
3. Evaluate (sin x + 1) dx.
0
R2
4. Prove the inequality (2 + x3 2x) dx > 8.
2
2R
5. Prove x(sin x + 1) dx < 2.
0
R
200
x+sin(x) dx 100 + 1 .
6. x 50
100
53
7. Denote by sn the area of
f(x; y) j 0 x 1; (1 x) log n + x ln(n + 1) y ln(1 + x)g
1
P
. Prove that sk < 1 .
k=1
2n
P k 2nP
+1 k
8. Prove that ( 1)k+1 xk < ln(1 + x) < ( 1)k+1 xk for x > 0.
k=1 k=1
9. Compute the logarithms of the primes 2,3,5,7 with a
ura
y 0:01.
10. Evaluate
R1 px dx.
0
R
11. Evaluate sin x dx.
0
54
2.3 Stieltjes Integral
On the
ontents of the le
ture. The Stieltjes relativization of integral
makes the integral
exible. We learn the main transformations of integrals.
They allow us to evaluate a lot of integrals.
55
Now let us estimate the dieren
es between integrals from (2.3.1) and their
approximations from (2.3.2). For example, for the right-hand side integrals one
has:
Zb Zb Zb Zb
(2.3.3) f d
g f" d
g = (f f" ) d
g " d
g = "(
g(b)
g(a)):
a a a a
Rb
Hen
e f d
g = I" + "1, where "1
"(g(b) g(a)). The same argument proves
a
Rb Rb
f dg = I" + "2 and
f dg = I" + "3 , where "2; "3
"(g(b) g(a)). Then the
a a
pairwise dieren
es between integrals of (2.3.1) do not ex
eed 2
"(g(b) g(a)).
Consequently they are less than any positive number, that is, they are zeros.
56
n Rb
P
is dened as sum of integrals fk dgk . Two dierential forms are
alled
k=1 a
equivalent on the interval [a; b℄ if their integrals are equal for all subintervals of
[a; b℄. For the sake of brevity we denote the dierential form f1 dg1 + f2 dg2 +
+ fn dgn by F dG, where F = ff1; : : : ; fng is a
olle
tion of integrands and
G = fg1; : : : ; gn g is a
olle
tion of dierands.
Theorem 4 (on multipli
ation). Let F dG and F 0 dG0 be two dierential
forms with positive in
reasing integrand and
ontinuous in
reasing dierands,
whi
h are equivalent on [a; b℄. Then their produ
ts by any in
reasing fun
tion f
on [a; b℄ are equivalent on [a; b℄ too.
Proof. If f is
onstant then the statement follows from the multipli
ation for-
mula. If f is pie
e-wise
onstant, then divide [a; b℄ into intervals where it is
onstant and prove the equality for parts and after
olle
t the results by the
Partition Rule. In the general
ase,
Zb Zb Zb Zb
0 fF dG [f ℄" F dG "F dG = " F dG:
a a a a
Rb Rb
Sin
e [f ℄" F 0 dG0 = [f ℄" F dG, one
on
ludes that
a a
Zb Zb Zb Zb
fF 0 dG0 fF dG " F dG + " F 0 dG0
a a a a
The right-hand side of this inequality
an be made arbitrarily small. Hen
e the
left-hand side is 0.
Integration by parts.
Theorem 5. If f and g are
ontinuous nonde
reasing nonnegative fun
tions
on [a; b℄ then d(fg) is equivalent to fdg + gdf .
Rd
Proof. Consider [
; d℄ [a; b℄. The integral f dg represents the area below the
Rd
urve (f (t); g(t))t2[
;d℄. And the integral g df represents the area on the left of
the same
urve. Its union is equal to [0; f (d)℄ [0; g(d)℄ n [0; f (
)℄ [0; g(
)℄. The
Rd
area of this union is equal to (f (d)g(d) f (
)g(
) = dfg. On the other hand
the area of this union is the sum of the areas of
urvilinear trapezia representing
Rd Rd
the integrals f dg and g df .
Rb
Change of variable Consider a Stieltjes integral f ( ) dg( ) and suppose
a
there is a
ontinuous nonde
reasing mapping : [t0 ; t1 ℄ ! [a; b℄, su
h that
57
(t0 ) = a and (t1 ) = b. The
omposition g( (t)) is a
ontinuous nonde
reasing
fun
tion and the
urve f(f ( (t); g( (t))) j t 2 [t0 ; t1 ℄g just
oin
ides with the
urve f(f ( ); g( )) j 2 [a; b℄. Hen
e, the following equality holds; it is known
as the Formula of Change of Variable:
Zt1 Z(t1 )
(2.3.6) f ( (t)) dg( (t)) = f ( ) dg( )
t0 (t0 )
For dierentials this means that the equality F (x)dG(x) = F 0 (x)dG0 (x)
on-
serves if one substitutes instead of an independent variable x a fun
tion.
Dierential Transformations
dxn . Integration by parts for f (t) = g(t) = t gives dt2 = tdt + tdt. Hen
e
2
tdt = t2 . If we already know that dxa = adxa 1 , then dxa+1 = d(xxa ) =
xdxa + xa dx = axxa 1 dx + xa dx = (a + 1)xa dx. This proves the Fermat
Theorem for natural a.
p p
d n x. To evaluate npd n x substitute x = npyn into equality dyn = nyn 1 dy. One
p
gets dx = nxdnpx x , hen
e d n x = n1 xdx
x .
ln xdx. We know d ln x = dxx . Integration by parts gives ln xdx = d(x ln x)
xd ln x = d(x ln x) dx = d(x ln x x).
Problems.
1. dx2=3
2. dx 1
3. x ln x dx
4. d ln2 x
5. ln2 x dx
6. dex
1
P 1
7. Investigate the
onvergen
e of k ln k .
k=1
58
2.4 Asymptoti
s of sums
On the
ontents of the le
ture. We be
ome at last a
quainted with the
fundamental
on
ept of limit. We extend the notion of sum of series and dis
over
that
hange of the order of summands
an ae
t the ultimate sum. Finally we
derive the famous Stirling formula for n!.
Asymptoti
formulas. Mer
ator series shows how useful series
an be for
evaluating of integrals. In this le
ture we will use integrals to evaluate both
partial and ultimate sums of series. Rarely one has an expli
it formula for
partial sums of a series. There are lots of important
ases where su
h a formula
does not exist. For example, it is known that partial sums of the Euler series
annot be expressed as a nite
ombination of elementary fun
tions. When an
expli
it formula is not available, one tries to nd a so-
alled asymptoti
formulas.
An asymptoti
formula for a partial sum Sn of a series is a formula of the type
Sn = f (n) + R(n) where f is a known fun
tion
alled prin
ipal part and R(n)
is a reminder, whi
h is small, in some sense, with respe
t to the prin
ipal part.
Today we will get an asymptoti
formula for partial sums of the harmoni
series.
59
Limit of sequen
e
Denition. A sequen
e fzk g of (
omplex) numbers
onverges to a number z if
lim z zk = 0. The number z is
alled limit of the sequen
e fzk g and denoted
by lim zk .
The innite sum represents a parti
ular
ase of a limit as demonstrate the
following.
P1
Theorem 2. The partial sums of an absolute
onvergent series zk
onverges
k=1
to its sum.
nP1 P1 1
P 1
P 1 1
Proof. j zk zk j = j zk j jzk j. Sin
e P jzk j > P jzk j ", there
k=1 k=1 k=n k=n k=1 k=1
nP1 P1
is a partial sum su
h that jzk j > jzk j ". Then for all m n one has
k=1 k=1
P1 P1
jzk j jzk j < ".
k=m k=n
60
Theorem 5 (integral test on
onvergen
e). If a nonnegative fun
tion f (x)
1
P R1
de
reases monotoni
ally on [1; +1), then f (k)
onverges i f (x) dx < 1.
k=1 1
R1 1 kR+1
P 1
P R1
Proof. Sin
e f (x) dx = f (x) dx, one has f (k) =
f + f (x) dx.
1 k=1 k k=1 1
1
P 1
Euler
onstant. The sum ln(1 + k1 ) , whi
h is
f for f (x) = x1 , is
k
k=1
alled Euler's
onstant and denoted by
. Its rst ten digits are 0:5772156649:::.
n
P
Harmoni
numbers The sum 1 is denoted Hn and is
alled the n-th
k
k=1
harmoni
number.
Theorem 6. Hn = ln n +
+ on where lim on = 0
nP1 nP1
Proof. Sin
e ln n = (ln(k + 1) ln k) = ln(1 + k1 ), one has ln n +
k=1 k=1
nP1 nP1
1 ln(1 + k1 ) = Hn 1 . But 1 ln(1 + k1 ) =
+ n , where lim n =
k k
k=1 k=1
0. Therefore Hn = ln n +
+ ( n1 + n ).
1 ( 1)k+1
P
Alternating harmoni
series The alternating harmoni
series k
k=1
is a
onditionally
onvergent series due to Leibniz's theorem, and it is not abso-
lutely
onvergent. To nd its sum we apply our Theorem 6 on the asymptoti
s
of harmoni
numbers.n
P ( 1)k+1
Denote by Sn = k the partial sum. Then Sn = Hn0 Hn00 , where
k=1
Pn n
0
Hn = 1 00 P 1 00 1 0
k [k is odd℄ and Hn = k [k is even℄. Sin
e H2n = 2 Hn and H2n =
k=1 k=1
H2n H200n = H2n 1 Hn one gets
2
1 1
(2.4.1) S2n = H2n H H = H2 n Hn =
2 n 2 n
ln 2n +
+ o2n ln n
on = ln 2 + (o2n on ):
n+1
Consequently Sn = ln 2 + (o[[n℄℄ o[n=2℄ + ( 1)n [n is odd℄). As the sum in
bra
kets is innitesimally small, one gets
X1 ( 1)n+1
= ln 2:
k=1 n
The same arguments for a permutated alternating harmoni
series give
1 1 1 1 1 1 1 1 3
(2.4.2) 1+ + +
3 2 5 7 4 9 11 6
+ + = log 2:
2
Indeed, in this
ase its 3n-th partial sum is S3n = H4n H200n = H4n 12 H2n
0
1 H = log 4n +
+ o 1 (log 2n +
+ o +log n +
+ o ) = log 4 1 log 2+ o0 =
n 4n 2n n n
32 log 2 + o0 , where lim o02 = 0. Sin
e the dieren
e between S and 2
S where
2 n n n 3 m
m = [n=3℄ is innitesimally small, this proves (2.4.2).
61
Stirling's Formula. We will try to estimate ln n!. The integration of inequal-
Rn
ities ln[x℄ ln x ln[x + 1℄ over [1; n℄ gives ln(n 1)! ln x dx ln n!. Let
1
Rn 1 (ln n! + ln(n
us estimate the dieren
e D between ln x dx and 1)!). 2
1
(2.4.3)
Zn nX1 Z1
1 p
D = (ln x (ln[x℄ + ln[x + 1℄)) dx = ln(k + x) ln k(k + 1) dx:
2 k=1
1 0
To prove that all summands on the left-hand side are nonnegative, we apply the
following general lemma.
R1 R1
Lemma 7. f (x) dx = f (1 x) dx for any fun
tion
0 0
Proof. The re
e
tion of the plane a
ross the line y = 21 transforms the
urvi-
linear trapezium of f (x) over [0; 1℄ into
urvilinear trapezium of f (1 x) over
[0; 1℄.
R1 p
Lemma 8. ln(k + x) dx ln k(k + 1)
0
Proof. Due to Lemma 7 one has
Z1 Z1 Z1
1
ln(k + x) dx = ln(k + 1 x) dx = (ln(k + x) + ln(k + 1 x)) dx =
2
0 0 0
Z1 Z1
p p
= ln (k + x)(k + 1 x) dx = ln k(k + 1) + x x2 dx
0 0
Z1
p p
ln k(k + 1) dx = ln k(k + 1)
0
62
1 1
P
We see that Dn 2k
1
2(k+1) = 1
2 for all n. Denote by D1 the sum
k=1
(2.4.3) for innite n. Then Rn = D1 Dn = 2n for some nonnegative < 1,
and we get
Zn Zn
1 ln n
(2.4.5) D1 = ln x dx (ln n! + ln(n 1)!) = ln x dx ln n! +
2n 2 2
1 1
Rn Rn
Substituting in (2.4.5) the value of the integral ln x dx = d(x ln x x) =
1 1
(n ln n n) (1 ln 1 1) = n ln n n + 1, one gets
ln n
(2.4.6) ln n! = n ln n n + + (1 D1 ) +
2 2n
Now we know that 1 (1 D1 ) 12 , but it is possible to evaluate
p the value
of D1 with more a
ura
y. Later we will prove that 1 D1 = 2.
Problems.
1
P
1. Does sin k
onverge?
k=1
1
P
2. Does sin k2
onverge?
k=1
3. Evaluate 1 + 21 + 41 + 51 62 +
2
3
2
3n + 3n1+1 + 3n1+2 : : :
1
P
4. If lim aann+1 < 1, then ak
onverge.
k=1
1
P
5. If jak ak 1 j < 1, then fak g
onverges.
k=1
p
( 1)[ k℄ 1
P
6. Prove
onvergen
e
k=1 k
1 1
P
7. Convergen
e ln3 k
k=1
1
P
8. Convergen
e p1
k=1 k ln k ln ln k
1
P 1
9. Convergen
e k ln k(ln ln k)2
k=1
1
P 1
10. Convergen
e k ln k and asymptoti
formula?
k=1
1
P 1
11. Convergen
e k ln2 k
k=1
63
12. Whi
h partial sum of above series is 0:01
lose to its ultimate sum?
1 1
P
13. Evaluate k ln2 k with pre
ision 0.01
k=1
R3
14. Evaluate ln x d[x℄
1
1 2n 2n
Q
15. Express the Stirling
onstant via the Wallis produ
t 2 =
n=1 2n 1 2n+1
64
2.5 Quadrature of
ir
le
On the
ontents of the le
ture. We extend the
on
eptHof the integral
to
omplex fun
tion. We evaluate a very important integral dzz , by apply-
ing Ar
himedes theorem on the area of
ir
ular se
tor. As a
onsequen
e, we
evaluate the Wallis produ
t and Stirling Constant.
Z Z Z Z Z
1
(2.5.2) dz =
os t d
os t + sin t d sin t i sin t d
os t + i
os t d sin t
z
p 0 0 0 0
R 1 2 R 1 2 R 1 2 2
Its real part transforms into 2 d
os t + 2 d sin t = 2 d(
os t + sin t) =
0 0 0
R 1
2 d1
= 0. An attentive reader has to obje
t: the integrals were dened only
0
for dierential forms with non-de
reasing dierand, while
os t de
reases.
Sign rule. Let us dene the integral for any dierential form fdg with any
ontinuous monotone dierand g and any integrand f of a
onstant sign (i.e,
non-positive or non-negative). The denition relies on the following Sign Rule
Zb Zb Zb
(2.5.3) f dg = f dg = f d( g )
a a a
65
If f is of
onstant sign, and g is monotone, then among the forms fdg, fdg,
fd( g) and fd( g) there is just one with non-negative integrand and non-
de
reasing dierand. For this form, the integral was dened earlier, for the other
ases it is dened by the Sign Rule.
Thus the integral of a negative fun
tion against an in
reasing dierand and
the integral of a positive fun
tion against a de
reasing dierand are negative.
And the integral of a negative fun
tion against a de
reasing dierand is positive.
Rb
The Sign Rule agrees with the Constant Rule: the formula
dg =
(g(b)
a
g(a)) remains true either for negative
or de
reasing g.
The Partition Rule also is not ae
ted by this extension of Integral.
The Inequality Rule takes the following form: if f1 (x) f2 (x) for all
Rb Rb
x 2 [a; b℄ then f1(x) dg(x) f2 (x) dg(x) for non-de
reasing dierand and
a a
Rb Rb
f1 (x) dg(x) f2 (x) dg(x) for non-in
reasing g.
a a
Change of variable. Now all integrals in (2.5.2) are dened. The next obje
-
tion
on
erns transformation
os td
os t = 21 d
os2 t. This transformation based
on a de
reasing
hange of variable x =
os t in dx2 =2 = xdx. But what happens
with an integral when one applies a de
reasing
hange of variable. The
urvi-
linear trapezium, whi
h represents the integral, does not
hange at all under
any
hange of variable, even for a non-monotone one. Hen
e the only thing
that may happen is a
hange of sign. And the sign
hanges by the Sign Rule,
simultaneously on the both sides of equality dx2 =2 = xdx. If the integrals of
xdx and dx2 were positive, both integrals of
os td
os t and
os2 t are negative
and have the same absolute value. These arguments work in the general
ase:
The de
reasing
hange of variable reverses the sign of the integral.
66
The integral of
os t d sin t is equal to the area of
urvilinear trapezium
OB 0 BA. The latter
onsists of a
ir
ular se
tor OBA with area =2 and a
R
triangle OB 0 B with area 1
os sin . Thus
os t d sin t = =2 + 1
os sin
2 2
R 1 0
As result we get z dz = i. This result has a lot of
onsequen
es. But
p
today we restri
t our attention to integrals of sin t and
os t.
B’ B
φ
O A’ A
67
Appli
ation of trigonometri
integrals.
Lemma 2. For any
onvergent innite produ
t of fa
tors 1 one has
n
Y 1
Y
(2.5.5) lim pk = pk
k=1 k=1
1
Q 1
Q
Proof. Let " be a positive number. Then pk > pk ", and by All-for-
k=1 k=1
n
Q 1
Q
One there is n su
h that pk > pk ". Then for any m > n one has the
k=1 k=1
1
Q m
Q Q1 m
Q Q1
inequalities pk pk > pk ". Therefore j pk p k j < ".
k=1 k=1 k=1 k=1 k=1
R R
Wallis produ
t. Set In = sinn x dx. Then I0 = 1 dx = and I1 =
0 0
R
sin x dx =
os +
os 0 = 2. For n 2, let us repla
e the integrand sinn x
0
by sinn 2 x(1
os2 x) and obtain
Z Z Z
In = sinn 2 x(1
os2 x) dx = sinn 2 x sinn 2 x
os x d sin x =
0 0 0
Z Z
1
= In 2
os x d sinn 1 (x) = I n 2 d(
os x sinn 1 x)+
n 1
0 0
Z
1
+ sinn 1 x d
os x = In I
2
n 1 n
0
68
Stirling
onstant. In the Le
ture 2.4 we have proved that
1
(2.5.9) ln n! = n ln n n + ln n + + on ;
2
where on is innitesimally small and is a
onstant. Now we are ready to
determine this
onstant. Consider the dieren
e ln 2n! 2 ln n! by (2.5.9) it
expands into (2n ln 2n 2n + 21 ln 2n + + o2n ) 2(n ln n n + 12 ln n + + on ) =
2n ln 2 + 21 ln 2n ln n + o0n , where o0n = o2n 2on is innitesimally small.
Then
an be presented as = 2 ln n! ln 2n!+2n ln 2+ 12 ln n + 21 ln 2 ln n + o0n .
Multiplying by 2 one gets 2 = 4 ln n! 2 ln 2n!+2 ln 22n ln n +ln 2+2o0n Hen
e
2 = (lim 4 ln n! 2 ln 2n! + 2 ln 22n ln n + ln 2) Swit
hing to the produ
ts and
keeping in mind identities n! = n!!(n 1)!! and n!2n = 2n!! one gets
(2.5.10)
n!4 24n+1 2 (2n!!)4 2 (2n!!)2 (2n + 1)
2 = lim 2 = lim 2 2 lim = 2
(2n!) n (2n!!) (2n 1)!! n (2n 1)!!(2n + 1)!!n
Problems.
R p
1. Evaluate 1 x2 dx
R
2. Evaluate p11 x2 dx
R p
3. Evaluate 5 x2 dx
R
4. Evaluate
os2 x dx
R
5. Evaluate tan x dx
R
6. Evaluate sin4 x dx
R
7. Evaluate sin x2 dx
R
8. Evaluate tan x dx
R
9. Evaluate x2 sin x dx
10. Evaluate d ar
sin
dx
x
R
11. Evaluate ar
sin x dx
R
12. Evaluate ex
os x dx
69
2.6 Virtually monotone fun
tions
Monotonization of integrand. Let us say that a pair of fun
tions f1 , f2
monotonize a fun
tion f , if f1 is a non-negative and non-de
reasing, f2 is a
non-positive and non-in
reasing and f = f1 + f2 .
Lemma 1. Let f = f1 + f2 and f = f10 + f20 be two monotonizations of f . Then
for any monotone h one has f1 dh + f2 dh = f10 dh + f20 dh.
Proof. Our equality is equivalent to f1 dh f20 dh = f10 dh f2 dh. By the sign
rule this turns into f1 dh + ( f20 ) dh = f10 dh + ( f2 ) dh. Now all integrands are
nonnegative and for non-de
reasing h we
an apply the Addition Theorem and
transform the inequality into (f1 f20 ) dh = (f10 f2 ) dh. This is true be
ause
(f1 f20 ) = (f10 f2 ).
The
ase of non-in
reasing dierand is redu
ed to the
ase of non-de
reasing
one by the transformation f1 d( h) + f2 d( h) = f10 d( h) + f20 d( h), whi
h is
based on the Sign Rule.
Lemma on lo
ally
onstant fun
tions. Let us say that a fun
tion f (x) is
lo
ally
onstant at a point x if f (y) = f (x) for all y suÆ
iently
lose to x, i.e
for all y from an interval (x "; x + ").
Lemma 3. A fun
tion f whi
h is lo
ally
onstant at any point of an interval
is
onstant.
Proof. Suppose f (x) is not
onstant on [a; b℄. We will
onstru
t by indu
tion a
sequen
e of intervals Ik = [ak ; bk ℄, su
h that I0 = [a; b℄, Ik+1 Ik , jbk ak j
2jbk+1 ak+1 j and the fun
tion f is not
onstant on ea
h Ik . First step. Let
= (a + b)=2, as f is not
onstant f (x) 6= f (
) for some x. Then
hoose [x;
℄ or
70
[
; x℄ as for [a1 ; b1 ℄. On this interval f is not
onstant. The same are all further
steps. The interse
tion of the sequen
e is a point su
h that any its neighborhood
ontains some interval of the sequen
e. Hen
e f is not lo
ally
onstant at this
point.
Lemma 4. If f (x) is a
ontinuous monotone fun
tion and a < f (x) < b then
a < f (y) < b for all y suÆ
iently
lose to x.
Proof. If f takes values greater than b, than it takes value b and if f (x) takes
values less than a then it takes value a due to
ontinuity. Then [f 1 (a); f 1 (b)℄
is the interval where inequalities hold.
Lemma 5. Let g1 , g2 be
ontinuous
omonotone fun
tions. Then g1 + g2 is
ontinuous and monotone, and for any virtually monotone f one has
(2.6.2) fdg1 + fdg2 = fd(g1 + g2 )
Proof. Suppose g1 (x) + g2(x) < p, let " = p g1 (x) g2 (x). Then g1 (y) <
g1 (y) + "=2 and g2(y) < g2(y) + "=2 for all y suÆ
iently
lose to x. Hen
e
g( y)+ g2 (y) < p for all y suÆ
iently
lose to x. The same is true for the opposite
inequality. Hen
e sgn(g1 (x) + g2 (x) p) is lo
ally
onstant at all points where
it is not 0. But it is not
onstant if p is an intermediate value, hen
e it is not
lo
ally
onstant, hen
e it takes value 0. At this point g1 (x) + g2 (x) = p and the
ontinuity of g1 + g2 is proved.
Consider a monotonization f = f1 + f2. Let gi be nonde
reasing. By
denition via monotonization of the integrand, the left-hand side of (2.6.2)
turns into (f1 dg1 + f2 dg1 )+(f1 dg2 + f2 dg2 ) = (f1 dg1 + f1 dg2 )+(f2 dg1 + f2 dg2 ).
By the Addition Theorem 3 f1 dg1 + f1 dg2 = f1 d(g1 + g2 ). And the equality
f2 dg1 + f2 dg2 = f2 d(g1 + g2 ) follows from ( f2 )dg1 + ( f2 )dg2 = ( f2 )d(g1 +
g2 ) by the Sign Rule. Hen
e the left-hand side is equal to f1 d(g1 + g2 ) +
f2 d(g1 + g2 ), whi
h
oin
ides with the right-hand side of (2.6.2) by denition
via monotonization of integrand. The
ase of non-in
reasing dierands is taken
are of via transformation of (2.6.2) by the Sign Rule into fd( g1)+ fd( g2) =
fd( g1 g2 ).
Lemma 6. Let g1 +g2 = g3 +g4 where all ( 1)k gk are non-in
reasing
ontinuous
fun
tions. Then fdg1 + fdg2 = fdg3 + fdg4 for any virtually monotone f
Proof. Our equality is equivalent to fdg1 fdg4 = fdg3 fdg2 . By the Sign
Rule it turns into fdg1 + fd( g4 ) = fdg3 + fd( g2). Now all dierands are
nonde
reasing and by Lemma 5 it transforms into fd(g1 g4 ) = fd(g3 g2 ).
This is true be
ause g1 g4 = g3 g2 .
71
Theorem 7 (Addition Theorem). For any virtually monotone fun
tions
f; f 0 and any virtually monotone
ontinuous g; g0 hold fdg + f 0 dg = (f + f 0 )dg
and fdg + fdg0 = fd(g + g0)
Proof. To prove fdg + f 0 dg = (f + f 0 )dg,
onsider a
ontinuous monotonization
g = g1 + g2. Then by denition of integral for virtually monotone dierand this
equality turns into (fdg1 + fdg2) + (f 0 dg1 + f 0 dg2 ) = (f + f 0 )dg1 + (f + f 0 )dg2 .
After rearranging it turns into (fdg1 + f 0dg1 ) + (fdg2 + f 0dg2 ) = (f + f 0 )dg1 +
(f + f 0 )dg2 But this is true due to Lemma 2.
To prove fdg + fdg0 = fd(g + g0 ),
onsider monotonizations g = g1 + g2 ,
g = g10 + g20 . Then (g1 + g10 ) + (g2 + g20 ) is monotonization for g + g0 . And
0
by the denition of integral for virtually monotone dierand our equality turns
into fdg1 + fdg2 + fdg10 + fdg20
Change of variable.
Lemma 8. If f is virtually monotone and g is monotone, then f (g(x)) is
virtually monotone.
Proof. Let f1 + f2 be a monotonization of f . If h is non-de
reasing then
f1 (h(x)) + f2 (h(x)) gives a monotonization of f (g(x)). If h is de
reasing the
monotonization is given by (f2 (h(x))+
)+(f1(h(x))
) where
is a suÆ
iently
large
onstant, to provide positivity of the rst bra
kets and negativity of the
se
ond one.
The following natural
onve
tion is applied to dene the integral with re-
versed limits.
Zb Za
f (x) dg(x) = f (x) dg(x)
a b
Theorem 9 (on
hange of variable). If h : [a; b℄ ! [h(a); h(b)℄ is monotone,
f (x) is virtually monotone, and g(x) is virtually monotone
ontinuous, then
Rb hR(b)
f (h(t)) dg(h(t)) = f (x) dg(x)
a h(a)
Proof. Let f = f1 + f2 and g = g1 + g2 be a monotonization and a
ontinu-
Rb
ous monotonization of f and g respe
tively. The f (h(t)) dg(h(t)) splits into
a
Rb
sum of four integrals. fi (h(t)) dgj (h(t)) where fi are of
onstant sign and gj
a
are monotone
ontinuous. This integrals
oin
ide with
orresponding integrals
hR(b)
fi (x) dgi (x). Indeed its absolute values are the area of the same
urvilinear
h(a)
trapezium. And their signs determined by the Sign Rule are the same.
72
(f +
)d(g +
)+(g +
)d(f +
). On the other hand d(f +
)(g +
) = dfg +
df +
dg
and (f +
)d(g +
) + (g +
)d(f +
) = fdg +
dg +
df . Compare these results
to get dfg = fdg + gdf . Now if f is in
reasing and g de
rease the g in
rease
and we get dfg = df ( g) = fd( g) + ( g)df = fdg gdf , whi
h lead to
dfg = fdg +gdf . The other
ases: f de
reasing, g in
reasing and both de
reasing
are proved by the same arguments. The extension of the Integration by Parts
to pie
e-wise monotone forms immediately follows by the Partition Rule.
73
Proof. Sin
e jf (xk+1 )+ g(xk+1 ) f (xk ) g(xk )j jf (xk+1 ) f (xk )j + jg(xk+1 )
g(xk )j, the variation of f + g along any sequen
e does not ex
eed the sum of the
variations of f and g along the sequen
e. Hen
e all partial variations of f + g
do not ex
eed varf [a; b℄ + varg [a; b℄, and so the same is true for the ultimate
variation.
Lemma 13. For any fun
tion of nite variation on [a; b℄, fun
tions varf [a; x℄
and varf [a; x℄ f (x) are both nonde
reasing fun
tions of x.
Proof. The nonde
reasing of varf [a; x℄ follows from nonnegativity and additivity
of variation. If x > y then the inequality varf [a; x℄ f (x) varf [a; y℄ f (y)
is equivalent to varf [a; x℄ varf [a; y℄ f (x) f (y). This is true be
ause
varf [a; x℄ varf [a; y℄ = varf [x; y℄ jf (x) f (y)j.
Lemma 14. varf 2 [a; b℄ 2(jf (a)j + varf [a; b℄) varf [a; b℄
Proof. For all x; y 2 [a; b℄ one has jf (x) + f (y)j = j2f (a) + f (x) f (a) +
f (y) f (a)j 2jf (a)j + varf [a; x℄ + varf [a; y℄ 2jf (a)j + 2 varf [a; b℄. Hen
e
nP1 nP1
jf 2 (xk+1 ) f 2 (xk )j = jf (xk+1 ) f (xk )jjf (xk+1 + f (xk )j 2(jf (a)j +
k=0 k=0
nP1
varf [a; b℄) jf (xk+1 ) f (xk )j 2(jf (a)j + varf [a; b℄) varf [a; b℄
k=0
Lemma 15. If varf [a; b℄ < 1 and varg [a; b℄ < 1, then varfg [a; b℄ < 1
Proof. 4fg = (f + g)2 (f g)2
Theorem 16. The fun
tion f is virtually monotone on [a; b℄ if and only if it
has a nite variation.
Proof. Sin
e monotone fun
tions have nite variation on nite intervals, and
the variation of a sum does not ex
eed the sum of variations one gets that
all virtually monotone fun
tions have nite variation. On the other hand if
f has a nite variation then f = (varf [a; x℄ +
) + (f (x) varf [a; x℄
), the
fun
tions in the bra
kets are monotone due to Lemma 13, and
hoosing
onstant
suÆ
iently large one obtains that the se
ond bra
ket is negative.
Problems.
Ri
1. Evaluate z 2 dz
1
2. Prove that 1=f (x) has nite variation if it is bounded.
Rb
3. Prove f (x) dg(x) max[a;b℄ f varg [a; b℄
a
74
Chapter 3
Derivatives
75
3.1 Newton-Leibniz Formula
Motivation. Consider the following problem: for a given fun
tion F nd a
Rd
fun
tion f su
h that dF (x) = f (x) dx, over [a; b℄, that is, f (t) dt = F (d) F (
)
for any subinterval [
; d℄ of [a; b℄.
Suppose that su
h an f exists. Sin
e the value of f at a single point does
not ae
ts the integrals, we
annot say anything about the value of f at any
given point. But if f is
ontinuous at a point x0 , its value is uniquely dened
by F .
To be pre
ise, the dieren
e quotient F (xx) Fx0(x0 ) tends to f (x0 ) as x tends to
Rx Rx
x0 . Indeed, F (x) = F (x0 )+ f (t) dt. Furthermore, f (t) dt = f (x0 )(x x0 )+
x0 x0
Rx Rx
(f (t) f (x0 )) dt. Also, j (f (t) f (x0 ) dtj varf [x0 ; x℄jx x0 j. Consequently
x0 x0
F (x) F (x0 )
(3.1.1) f (x0 ) varf [x; x0 ℄
x x0
However, varf [x; x0 ℄
an be made arbitrarily small by
hoosing x suÆ
iently
lose to x0 , sin
e varf x0 = 0.
76
Proof. This follows immediately from Lemma 1.
Lemma 4. If the limits xlim
!x f (x) and xlim !x f (x)g(x)
!x g(x) exist, then also xlim
0 0 0
!x f (x)g(x) = xlim
and xlim
0 0 !x g(x) exist.
!x f (x) xlim 0
Proof. If f (x) = A + o(x) and g(x) = B + !(x), then f (x)g(x) = AB + A!(x) +
Bo(x)+ !(x)o(x), where A!(x), Bo(x) and !(x)o(x) all are innitesimally small
at x0 by Lemma 2, and their sum is innitesimally small by Lemma 1.
Denition. A fun
tion f is
alled
ontinuous at x0 , if xlim
!x f (x) = f (x0 ).
0
A fun
tion is said to be
ontinuous (without mentioning a point), if it is
ontinuous at all points under
onsideration.
The following lemma gives a lot of examples of
ontinuous fun
tions.
Lemma 5. If f is a monotone fun
tion on [a; b℄ su
h that f [a; b℄ = [f (a); f (b)℄
then f is
ontinuous.
Proof. Given a positive ". Suppose f is nonde
reasing. For a given point x
denote by x" = f 1 (f (x) + ") and x" = f 1 (f (x) "). Then [x" ; x" ℄
ontains a
neighborhood of x, and for any y 2 [x" ; x" ℄ one has f (x) + " = f (x" ) f (y)
f (x" ) = f (x) + ". Hen
e the inequality jf (y) f (x)j < " holds lo
ally at x for
any ".
The following theorem immediately follows from the Corollary 3 and Lemma
4.
Theorem 6. If the fun
tions f and g are
ontinuous at x0 , then f + g and fg
are
ontinuous at x0 .
The following property of
ontinuous fun
tion is very important.
Theorem 7. If f is
ontinuous at x0 and g is
ontinuous at f (x0 ), then g(f (x))
is
ontinuous at x0 .
Proof. Given " > 0, we have to nd a neighborhood U of x0 , su
h that jg(f (x))
g(f (x0 ))j < " for x 2 U . As lim g(y) = g(f (x0 )), there exists a neighborhood
y!f (x0 )
V of f (x0 ) su
h that jg(y) g(y0)j < " for y 2 V . Thus it is suÆ
ient to nd a U
su
h that f (U ) V . And we
an do this. Indeed, by denition of neighborhood
there is Æ > 0, su
h that V
ontains VÆ = fy j jy f (x0)j < Æg. Sin
e xlim
!x0 f (x) =
f (x0 ), there is a neighborhood U of x0 su
h that jf (x) f (x0 )j < Æ for all x 2 U .
Then f (U ) VÆ V .
Denition. A fun
tion f is
alled dierentiable at a point x0 if the dieren
e
quotient f (xx) x(0f0 ) has a limit as x tends to x0 . This limit is
alled the derivative
of the fun
tion F at the point x0 , and denoted f 0 (x0 ) = xlim f (x) f (x0 )
!x x x0
0
Immediately from the denition one evaluates the derivative of linear fun
-
tion.
(3.1.2) (ax + b)0 = a
The following lemma is a dire
t
onsequen
e of Lemma 3.
Lemma 8. If f and g are dierentiable at x0 , then f + g is dierentiable at x0
and (f + g)0 (x0 ) = f 0 (x0 ) + g0 (x0 )
77
Linearization. Let f be dierentiable at x0 . Denote by o(x) the dieren
e
f (x) f (x0 )
x x0 f 0(x0 ). Then
(3.1.3) f (x) = f (x0 ) + f 0 (x0 )(x x0 ) + o(x)(x x0 );
where o(x) is innitesimally small at x0 . We will
all su
h a representation a
linearization of f (x).
Lemma 9. If f is dierentiable at x0 , then it is
ontinuous at x0 .
Proof. All summands but f (x0 ) on the right-hand side of (3.1.3) are innitesi-
!x f (x) = f (x0 ).
mally small at x0 ; hen
e xlim
0
Lemma 10 (on uniqueness of linearization). If f (x) = a + b(x x0 ) +
!x0 o(x) = 0, then f is dierentiable and at x0 , a = f (x0 )
o(x)(x x0 ), where xlim
0
and b = f (x0 ).
Proof. The dieren
e f (x) f (x0 ) is innitesimally small at x0 , be
ause f
is
ontinuous at x0 , and the dieren
e f (x) a = b(x x0 ) + o(x)(x x0 )
is innitesimally small by the denition of linearization. Hen
e f (x0 ) a is
innitesimally small. But it is
onstant, hen
e f (x0 ) a = 0.
Thus we established a = f (x0 ). The dieren
e fx(x)x0a b = o(x) is innites-
imally small as well as f (xx) xf 0(x0) f 0 (x0 ). But f (xx) xf 0(x0) = fx(x)x0a . Therefore
b f 0 (x0 ) is innitesimally small. That is b = f 0 (x0 ).
Lemma 11. If f and g are dierentiable at x0 , then fg is dierentiable at x0
and (fg)0 (x0 ) = f 0 (x0 )g(x0 ) + g0 (x0 )f (x0 )
Proof. Consider lineariations f (x0 ) + f 0(x0 )(x x0 ) + o(x)(x x0 ) and g(x0 ) +
g0 (x0 )(x x0 ) + !(x)(x x0 ). Their produ
t is f (x0 )g(x0 ) + (f 0 (x0 )g(x0 ) +
f (x0 )g0 (x0 ))(x x0 ) + (f (x)!(x) + f (x0 )o(x))(x x0 ). This is the linearization
of f (x)g(x) at x0 , be
ause f! and go are innitesimally small at x0 .
Theorem 12. If f is dierentiable at x0 , and g is dierentiable at f (x0 ) then
g(f (x)) is dierentiable at x0 and (g(f (x0 )))0 = g0 (f (x0 ))f 0 (x0 )
Proof. Denote f (x0 ) by y0 and substitute into the linearization g(y) = g(y0 ) +
g0 (y0 )(y y0 ) + o(y)(y y0 ) another linearization y = f (x0 ) + f 0 (x0 )(x x0 ) +
!(x)(x x0 ). Sin
e y y0 = f 0 (x0 )(x x0 ) + !(x)(x x0 ), we get g(y) =
g(y0 ) + g0 (y0 )f 0 (x0 )(x x0 ) + g0 (y0 )(x x0 )!(x) + o(f (x))(x x0 ). Due to
the Lemma on linearization, it is suÆ
ient to prove that g0 (y0 )!(x) + o(f (x))
is innitesimally small at x0 . The rst summand is obviously innitesimally
small. To prove that the se
ond one also is innitesimally small, we remark
that o(f (x0 ) = 0 and o(y) is
ontinuous at f (x0 ) and that f (x) is
ontinuous at
x0 due to Lemma 9. Hen
e by Theorem 3.1 the
omposition is
ontinuous at
x0 and innitesimally small.
Theorem 13. Let f be a virtually monotone fun
tion on [a; b℄. Then F (x) =
Rx
f (t) dt is virtually monotone and
ontinuous on [a; b℄. It is dierentiable at
a
any point x0 where f is
ontinuous, and F 0 (x0 ) = f (x0 ).
78
Proof. If f has a
onstant sign, then F is monotone. So, if f = f1 + f2 is a
Rx Rx
monotonization of f , then f1 (x) dx + f1 (x) dx is a monotonization of F (x).
a a
This proves that F (x) is virtually monotone.
To prove
ontinuity of F (x) at x0 , x a
onstant C whi
h bounds f in some
neighborhood U of x0 . Then for x 2 U one proves that jF (x) F (x0 )j is
Rx Rx
innitesimally small via inequalities jF (x) F (x0 )j = j f (x) dxj j C dxj =
x0 x0
C jx x0 j.
Now suppose f is
ontinuous at x0 . Then o(x) = f (x0 ) f (x) is innites-
x
imally small at x0 . Therefore xlim 1 R o(x) dx = 0. Indeed for any " > 0
!x x x00 x0
Rx
the inequality jo(x)j " holds over [x" ; x0 ℄ for some x" . Hen
e j o(x) dxj
x0
Rx
j " dxj = "jx x0 j for any x 2 [x0 ; x" ℄.
x0
Rx
Then F (x) = F (x0 )+f (x0 )(x x0 )+( x 1x0 o(t) dt)(x x0 ) is a linearization
x0
of F (x) at x0 .
Corrolary 14. The fun
tions ln, sin,
os are dierentiable and ln0 (x) = x1 ,
sin0 =
os,
os0 = sin.
Proof. Sin
e d sin x =
os x dx, d
os x = sin x dx, due to Theorem 13 both
sin x and
os x are
ontinuous, and, as they are
ontinuous, the result follows
from Theorem 13. And ln0 x = x1 , by Theorem 13, follows from
ontinuity of
1
x . The
ontinuity follows from Lemma 5.
Sin
e sin0 (0) =
os 0 = 1 and sin 0 = 0, the linearization of sin x at 0 is
x + xo(x). This implies the following very important equality
sin x
(3.1.4) lim
x!0
=1
x
Lemma 15. If f 0 (x) > 0 for all x 2 [a; b℄, then f (b) > f (a)j
Proof. Suppose f (a) f (b). We
onstru
t a sequen
e of intervals [a; b℄
[a1 ; b1 ℄ [a2 ; b2 ℄ : : : . su
h that their lengths tend to 0 and f (ak ) f (bk ).
All steps of
onstru
tion are the same. The general step is: let m be the middle
point of [ak ; bk ℄. If f (m) f (ak ) we set [ak+1 ; bk+1 ℄ = [ak ; m℄, otherwise
f (m) > f (ak ) f (bk ) and we set [ak+1 ; bk+1 ℄ = [m; bk ℄.
Now
onsider a point x belonging to all [ak ; bk ℄. Let f (y) = f (x) + (f 0 (x) +
o(x))(y x) be the linearization of f at x. Let U be neighborhood where
jo(x)j < f 0(x). Then sgn(f (y) f (x)) = sgn(y x) for all y 2 U . However for
some n we get [an ; bn ℄ U . If an x < bn we get f (an ) f (x) < f (bn ) else
an < x and f (an ) < f (x) f (bn ). In the both
ases we get f (an ) < f (bn ).
This is
ontradi
tion with our
onstru
tion of the sequen
e of intervals.
Theorem 16. If f 0 (x) = 0 for all x 2 [a; b℄, then f (x) is
onstant.
79
Proof. Set k = f (bb) fa(a) . If k < 0 then g(x) = f (x) kx=2 has derivative
g0 (x) = f 0 (x) k=2 > 0 for all x. Hen
e by Lemma 15 g(b) > g(a) further
f (b) f (a) > k(b a)=2. This
ontradi
t to denition of k. If k > 0 then the
same
ontradi
tion one gets
onsidering g(x) = f (x) + kx=2.
Theorem 17 (Newton-Leibniz). If f 0 (x) is a
ontinuous virtually monotone
Rb
fun
tion on an interval [a; b℄, then f 0 (x) dx = f (b) f (a)
a
R x
Proof. Due to Theorem 13, the derivative of the dieren
e f 0(t) dt f (x) is
a
zero. Hen
e the dieren
e is
onstant by Theorem 16. Substituting x = a we
Rx
nd the
onstant whi
h is f (a). Consequently, f 0(t) dt f (x) = f (a) for all
a
x. In parti
ular, for x = b we get the Newton-Leibniz formula.
Problems.
p p
1. (1=x)0 , x0 , ( sin x2 )0
2. exp0 x
3. ar
tg0 x, tan0 x
4. jxj0 , Re z 0
5. f 0(x) 1 () f (x) = x +
onst
!0
x2
R sin t
6. t dt
x x
p 0
7. 1 x2
1 0
R sin kt
8. t dt
0 k
9. If f is
ontinuous at a and nlim
!1 xn = a then nlim
!1 f (xn ) = f (a).
y 0
R
10. [x℄ dx .
0 y
11. ar
sin0 x
R dx
12. 2+3x2
13. If f 0(x) < 0 for all x < m and f 0 (x) > 0 for all x > m then f 0 (m) = 0
14. If f 0(x) is bounded on [a; b℄ then f is virtually monotone
80
3.2 Exponential Fun
tion.
On the
ontents of the le
ture. We solve the prin
ipal dierential equa-
tion y0 = y. Its solution, the exponential fun
tion, is expanded into a power
series. We be
ome a
quainted with hyperboli
fun
tions. And, nally, we prove
irrationality of e.
P P’
T T’ V V’
a b
81
First,
onsider the
ase of a horizontal tangent line. In this
ase f (x0 ) is
either maximal or minimal value of f (x).
Lemma 1. If a fun
tion f (x) is dierentiable at an extremal point x0 , then
f 0 (x0 ) = 0.
Proof. Consider the linearization f (x) = f (x0 ) + f 0(x0 )(x x0 ) + o(x))(x
x0 ). Denote x x0 by x, and f (x) f (x0 ) by f (x). If we suppose that
f 0 (x0 ) 6= 0, then, for suÆ
iently small x, we get jo(x x)j < jf 0 (x)j, hen
e
sgn(f 0 (x0 ) + o(x0 + x)) = sgn(f 0 (x0 ) + o(x0 x)), and sgn f (x) = sgn x.
Therefore the sign of f (x)
hanges whenever the sign of x
hanges. The
sign of f (x)
annot be positive, if f (x0 ) is the maximal value of f (x), and it
annot be negative, if f (x0 ) is the minimal value. This is the
ontradi
tion.
Theorem 2. If a fun
tion f (x) is dierentiable at x0 and its graph is
onvex,
then a tangent line to the graph of f (x) at x0 is y = f (x0 ) + f 0 (x0 )(x x0 ).
Proof. Let y = ax + b be the equation of a tangent line to the graph y = f (x) at
the point x0 . Sin
e ax + b passes through x0 , one has ax0 + b = f (x0 ), therefore
b = f (x0 ) ax0 , and it remains to prove that a = f 0 (x0 ). If the tangent line
ax + b is not horizontal,
onsider the fun
tion g(x) = f (x) ax. At x0 it takes
either a maximal or a minimal value and g0 (x0 ) = 0 by Lemma 1. On the other
hand, g0 (x0 ) = f 0 (x0 ) a.
Exponenta. The fun
tion inverse to the natural logarithm is
alled the expo-
nential fun
tion. We shall
all it exponenta to distinguish it from other expo-
nential fun
tions.
Theorem 3. The exponenta is the unique solution of the dierential equation
y0 = y su
h that y0 (0) = 1.
x 0
Proof. Dierentiation of equality ln exp x = x gives expexp x = 1. Hen
e exp x
satises the dierential equation y = y. For x = 0 this equation gives exp0 (0) =
0
exp 0. But exp 0 = 1 as ln 1 = 0.
For the
onverse, let y(x) be a solution of y0 = y. The derivative of ln y is
0
y = 1. Hen
e the derivative of ln y (x) x is zero. By Theorem 16 from the
y
previous le
ture, this implies ln y(x) x =
for some
onstant
. If y0 (0) =
1, then y(0) = 1 and
= ln 1 0 = 0. Therefore ln y(x) = x and y(x) =
exp ln y(x) = exp x.
82
Exponential series. Our next goal is to prove that
x2 x3 xk X1 xn
(3.2.2) exp x = 1 + x + + + + + = ;
2 23 k! k=0
n!
where 0! = 1. This series is absolutely
onvergent for any x. Indeed, the ratio
of its subsequent terms is nx tends to 0, hen
e it is eventually majorized by any
geometri
series.
Hyperboli
fun
tions. To prove that the fun
tion presented by series (3.2.2)
is virtually monotone,
onsider its odd and even parts. These partes represent
so
alled hyperboli
fun
tions : hyperboli
sine sh x, and
osine
h x.
1
X x2k+1 1
X x 2k
(3.2.3) sh(x) =
h(x) =
k=0 (2k + 1)! k=0 (2k )!
The hyperboli
sine is an in
reasing fun
tion, as all odd powers are in
reasing
over the whole line. The hyperboli
osine is in
reasing for positive x and
de
reasing for negative. Hen
e both are virtually monotone; and so is their
sum.
Rx
Consider the integral sh t dt. As all terms of the series representing sh are
0
in
reasing, we
an integrate the series termwise. This integration gives
h x. As
sh x is lo
ally bounded,
h x is
ontinuous by Theorem 13. Consider the integral
Rx
h t dt, here we also
an integrate the series representing
h termwise, be
ause
0
for positive x all the terms are in
reasing, and for negative x, de
reasing. The
integration gives sh x. Sin
e the
ontinuity of
h x was already proved. Further,
by Theorem 13 we get that sh x is dierentiable and sh0 x =
h x. Now returning
Rx
to equality
h x = sh t dt we get
h0 x = sh x, as sh x is
ontinuous.
0
Therefore (sh x +
h x)0 =
h x + sh x. And sh 0 +
h 0 = 0 + 1 = 1. Now by
the above theorem 3 one gets exp x =
h x + sh x.
Other exponential fun
tions. The exponenta as fun
tion inverse to loga-
rithm transforms sums into produ
ts. That is for all x and y one has
(3.2.4) exp(x + y) = exp x exp y
A fun
tion whi
h has this property (i.e., transform sums into produ
ts) is
alled
exponential.
Theorem 4. For any positive a there is a unique dierentiable fun
tion denoted
by ax
alled exponential fun
tion to base a, su
h that a1 = a and ax+y = axay
for any x, y. This fun
tion is dened by the formula exp a ln x.
Proof. Consider l(x) = ln ax. This fun
tion has the property l(x + y) = l(x) +
l(y). Therefore its derivative at any point is the same: it is equal to k = xlim l(x)
!0 x .
Hen
e the fun
tion l(x) kx is
onstant, be
ause its derivative is 0. This
onstant is equal to l(0), whi
h is 0. Indeed l(0) = l(0 + 0) = l(0) + l(0). Thus
83
ln ax = kx. Substituting x = 1 one gets k = ln a. Hen
e ax = exp(x ln a).
So if a dierentiable exponential fun
tion with base a exists, it
oin
ides with
exp(x ln a). On the other hand it is easy to see that exp(x ln a) satises all
requirements for an exponential fun
tion to base a, that is exp(1 ln a) = a,
exp((x + y) ln a) = exp(x ln a) exp(y ln a); and it is dierentiable as
omposition
of dierentiable fun
tions.
Powers. Hen
e for any positive a and any real b one denes the number ab as
ab = exp(b ln a)
a is
alled base, and b is
alled exponent. For rational b this denition agrees
with the old denition. Indeed if b = pq then properties of exponent and loga-
p p
rithm imply a q = q ap .
Earlier, we have dened logarithms to base b as the number
,
alled loga-
rithm of b to base a, if ab =
and denoted
= loga b.
The basi
properties of powers are
olle
ted here.
Theorem 5.
log a
(ab ) = a(b
) ab+
= ab a
(ab)
= a
b
loga b =
log b
Power fun
tions The power operation allows us to dene power fun
tion x
for any real degree . Now we
an prove the equality (x )0 = x 1 in its full
value. Indeed, (x )0 = (exp( ln x))0 = exp0 ( ln x)( ln x)0 = exp( ln x) x =
x 1
1
Y 1
X
(3.2.5) ln pk = ln pk
k=1 k=1
Proof.
1
X n
X n
X
exp( ln pk ) = exp(nlim
!1 log pk ) = nlim
!1 exp( log pk ) =
k=1 k=1 k=1
Yn Y1
lim
n!1
pk = pk :
k=1 k=1
84
1
ln pk ) = 1
P Q
Hen
e exp( k=1 pk . Taking logarithm of both sides one gets (3.2.5)
k=1
1
X 1
Y
(3.2.6) exp ak = exp ak
k=1 k=1
Problems.
1. Prove the inequalities 1 + x exp x 1
1 x
2. Prove the inequalities 1+x x log(1 + x) x
3. Evaluate nlim 1 n
!1 1 n
4. Evaluate nlim 2 n
!1 1 + n
5. Evaluate nlim 1 n
!1 1 + n2
6. Find the derivative of xx .
7. x > y implies exp x > exp y
8. Express via e: exp 2, exp(1=2), exp(2=3), exp( 1)
m
9. Prove that exp(m=n) = e n
10. Prove that exp x > 0 for any x.
11. Prove the addition formulas
h(x + y) =
h(x)
h(y)+sh(x) sh(y) sh(x + y) = sh(x)
h(y)+sh(y)
h(x)
85
12. Prove that sh(x 0:5) = sh 0:5
h(x)
h(x 0:5) = sh 0:5 sh(x).
13. Prove sh 2x = 2 sh x
h x
14. Prove
h2 (x) sh2 (x) = 1
15. Solve the equation sh x = 4=5
P1
16. Express via e the sum k=k!
k=1
1
P
17. Express via e the sum k2 =k!
k=1
18. Prove that f exp
kn g is unbounded
k
Q P
19. The produ
t (1 + pn )
onverges i the sum pn (pn 0)
onverges.
Q e1=n
20. Determine
onvergen
e
1 + n1
Q
21. Does n(e1=n 1)
onverges?
1 [k prime℄
P
22. Prove divergen
e of k .
k=1
23. Expand ax into a power series
24. Geometri
al sense of sh x and
h x
!1 sin en!
25. Evaluate nlim
1
P
26. Does the series sin en!
onverge?
k=1
27. Prove irrationality of e2
86
3.3 Euler Formula
Complex Newton-Leibniz. For a fun
tion of a
omplex variable f (z ) the
derivative is dened by the same formula f 0 (z0 ) = zlim f (z) f (z0 )
!z0 z z0 . We will
denote it also by dfdz(z) , to distinguish from derivative of pathes:
omplex valued
fun
tions of real variable. For a path p(t) its derivative will be denoted either
p0 (t) or dpdt(t) . The Newton-Leibniz formula for real fun
tions
an be expressed
The equality dfdt(t) dt = df (t). Now we extend this formula to
omplex fun
tions.
The linearization of a
omplex fun
tion f (z ) at z0 has the same form f (z0)+
f 0 (z0 )(z z0 ) + o(z )(z z0 ), where o(z ) is an innitesimally small fun
tion of
omplex variable. The same arguments as for real numbers prove the basi
rules
of dierentiation: the derivative of sum, produ
t and
omposition.
Theorem 1. dzn = nz n 1
dz
Proof. dz
dz = 1 one gets immediately from the denition of the derivative. Sup-
n n+1 n dzn
dz
pose the equality dz = nz n 1 is proved for n. Then dzdz = dzz dz = z dz +
z n dz
dz = znz
n 1 + z n = (n +1)z n dz . And the theorem is proved by indu
tion.
87
1
P
Dierentiation of series. Let us say that a series of
omplex numbers ak
k=1
P1
majorizes (eventually) another su
h series bk if jbk j jak j for all k (resp. for
k=1
k onward some n).
P1 P1
The series k
k (z z0)k 1 is
alled a formal derivative of
k (z z0)k .
k=1 k=0
1
P
Lemma 5. Any power series
k (z z0 )k eventually majorizes its formal
k=0
1
P
derivative k
k (z1 z0 )k 1 if jz1 z0 j < jz z0 j.
k=0
Proof. The ratio of n-th term of the derivative to the n-th term of the series
k
tends to 0 as n tends to innity. Indeed, this ratio is k((zz1 z0z)0k) = kqk , where
jqj < 1 sin
e jz1 z0 j < jz z0. The fa
t that nlim !1 nqn = 0 follows from
P1 k
onvergen
e of kq whi
h we already have proved before. This series is
k=1
P1
eventually majorized by any geometri
series AQk with Q > q.
k=0
A path p(t) is
alled monotone if both Re p(t) and Im p(t) are monotone.
Lemma 6. Let p : [a; b℄ ! C be a smooth monotone path, and let f (z ) be
R
virtually monotone. If jf (p(t))j
for t 2 [a; b℄ then f (z ) dz 4
jp(b)
p
p(a)j
Proof. Integration
of the inequalities
Re f (p(t))
against d Re z along
R
the path gives Re f (z ) d Rez
j Re p(b) Re p(a)j
jp(b) p(a)j. The same
p
R
arguments prove Im f (z ) d Imz
Im p(b)
p
j Im p(a)j
jp(b) p(a)j. The
R
sum of these inequalities gives Re f (z ) dz
p
2
j Re p(b) Re p(a)j. The same
R
arguments yields Im f (z ) dz 2
Re p(b)
j Re p(a)j. And the addition of
p
the two last
inequalities
allow
us to a
omplish
the proof of the Lemma be
ause
R R R
f (z ) dz
p
Re f (z ) dz + f (z ) dz.
p p
Lemma 7. jz n n j njz j maxfjz n 1j; j n 1 jg
nP1
Proof. (z n n ) = (z ) zk n k 1 and jz k n k 1 j maxfjz n 1 j; j n 1 jg.
k=0
88
Rb
We denote by f (z ) dz the integral along the linear path from a to b.
a
Lemma 8. For any
omplex z , and natural n > 0 one has
(3.3.1) jz n z0n nz0n 1(z z0 )j 2n(n 1)jz z0 j2 maxfjz jn 2; jz0 jn 2g
Rz
Proof. By the Newton-Leibniz formula, z n z0n = n n 1 d . Further,
z0
Zz Zz Zz
n n 1 d = nz0n 1 d + n( n 1 z0n 1) d =
z0 z0 z0
Zz
nz0n 1+ n( n 1 z0n 1) d:
z0
Rz
Consequently, the left-hand side of (3.3.1) is equal to n( n 1 z0n 1 ) d .
z0
Due to Lemma 7 the absolute value of the integrand along the linear path does
not ex
eed (n 1)jz z0 j maxfjz n 2j; jz0n 2 jg. Now the estimation of the integral
by Lemma 6 gives just the inequality (3.3.1).
P1 1
P
Theorem 9. If
k (z1 z0 )k
onverges absolutely, then
k (z z0 )k and
k=0 k=0
1
P
k
k (z z0 ) absolutely
onverge provided by jz z0j < jz1 z0j, and the
k 1
k=1
1
P 1
P
fun
tion k
k (z z0 )k 1 is the
omplex derivative of
k (z z0 )k .
k=1 k=0
1
P
Proof. Series
k (z z0 )k and its formal derivative are eventually majorized
k=0
1
P
by
k (z1 z0 )k if jz z0j jz1 z0 j by the Lemma 5. Hen
e they absolutely
k=0
onverge in the
ir
le jz z0 j jz1 z0 j. Consider
X1 1
X X1
(3.3.2) R(z ) =
k (z z0 )k
k ( z0 ) (z ) k
k ( z0 )k 1 :
k
k=0 k=0 k=1
1
P
To prove that the formal derivative is the derivative of
k (z z0 )k at it is
k=0
suÆ
ient to prove that R(z ) = o(z )(z ), where o(z ) is innitesimally small at
P1
. One has R(z ) =
k (z z0 )k ( z0 )k k( z0 )k 1 . By Lemma 8
k=1
P1
one gets the following estimate: jR(z )j 2j
k jk(k 1)jz j2 jz2 z0 jn 2 ,
k=1
where jz2 z0 j = maxfjz z0 j; j z0 jg. Hen
e all we need now is to prove
P1
that 2k(k 1)j
k jjz2 z0jk 2 jz j is innitesimally small at . And this in
k=1
1
P
its turn follows from the
onvergen
e of 2k(k 1)j
k jjz2 z0 jk 2 . The later
k=1
89
may be dedu
ed from the Lemma 5. Indeed,
onsider z3 , su
h that jz2 z0 j <
1
jz3 z0 j < jz1 z0 j. The
onvergen
e of P kj
k jjz3 z0 jk 1 follows from
k=1
1
P
the
onvergen
e of j
k jjz1 z0 j by the Lemma 5. And
onvergen
e of
k
k=0
1
P P1
k(k 1)j
k jjz2 z0 jk 2 follows from
onvergen
e of kj
k jjz3 z0 jk 1 by
k=2 k=1
the same Lemma
P1
Corrolary 10. Let f (z ) =
k z k absolutely
onverges for jz j < r, and a; b
k=0
Rb 1
k+1 k+1
P
have absolute values less then r. Then f (z ) dz = k+1 (b
k a )
a k=0
P1
zk+1
Proof. Consider F (z ) = k+1 . This series is term-wise majorized by the
k
k=0
series of f (z ), hen
e it
onverges absolutely for jz j < r. By the Theorem 9 f (z )
is its derivative for jz j < r. In our
ase f (z ) is dierentiable and its derivative
P1
is bounded by kj
k jr0k , where r0 = maxfjaj; jbjg. Hen
e f (z ) is
ontinuous
k=0
and virtually monotone and our result now follows from the Theorem 3.
90
termwise dierentiation. Su
h dierentiation simply gives the original series,
whi
h represents z1 in this
ir
le.
Rz
Consider the fun
tion l(z ) = 1 d . Then l0 (z ) = z1 . The derivative of the
z0
omposition exp l(z ) is expzl(z) . Hen
e the
omposition satises the dierential
equation y0 z = y. We sear
h for a solution of this equation in the form y = zw.
Then y0 = w + w0 z and our equation turns into wz + w0 z 2 = wz . Therefore
w0 = 0 and w is
onstant. To nd this
onstant substitute z = z0 and get
1 = exp 0 = exp l(z0) = wz0 . Hen
e w = z10 and exp l(z ) = zz0 .
To prove the general
ase
onsider a partition fxk gnk=0 of [a; b℄. Denote by
pk the restri
tion of p over [xk ; xk+1 ℄. Choose the partition so small that jp(x)
p(xk )j < jp(xk )j for all x 2 [xk ; xk+1 ℄. Then
R 1
any pk satises the requirement
R 1
of
p (x k+1 )
the above
onsidered
ase. Hen
e exp d = p(xk ) . Further exp d =
pk p
nP1 R Qn 1 p(xk+1 )
1
exp d = k=0 p(xk ) = p(b)=p(a).
k=0 pk
Theorem 13 (Euler Formula). For any real one has
(3.3.4) exp i =
os + i sin
R 1
Proof. In Le
ture 2.5 we have evaluated z dz = i for p(t) =
os t + i sin t, t 2
p
[0; ℄. Hen
e Lemma 12 applied to p(t) immediately gives the Euler formula.
Trigonometri
fun
tions in C . The Euler formula gives power series ex-
pansions for sin x and
os x
X1 x2k+1 X1 x2k
(3.3.5) sin x = ( 1)k
os x = ( 1)k
k=0
(2k + 1)! k=0
(2k)!
These expansions are used to dene the trigonometri
fun
tions for
omplex
variable. On the other hand the Euler formula allows us to express trigonometri
fun
tions via the exponenta:
exp(iz ) exp( iz ) exp(iz ) + exp( iz )
(3.3.6) sin z =
os z =
2i 2
The other trigonometri
fun
tions tan,
ot, se
,
ose
are dened for
omplex
variable by usual formulas via sin and
os.
Problems.
1 sin k
P
1. Evaluate k!
k=1
R1 1 ( 1)k+1
P
2. Prove the formula of Joh. Bernoulli xx dx = kk
0 k=1
3. Find ln( 1).
91
4. Solve the equation exp z = i.
5. Evaluate ii .
eize iz eiz + e iz
6. Prove sin z =
os z =
2i 2
2 2
7. Prove the identity sin z +
os z = 1
8. Solve the equation sin z = 5=3
1 sin n
P
9. Evaluate n!
k=1
10. Solve the equation
os z = 2
1
os n
P
11. Evaluate n!
k=0
H dz
12. Evaluate
jzj=1 z2
1 k sin kx
P
13. Evaluate q
k=1 k
14. Expand into a power series ex
os x
92
3.4 Abel's Theorem
On the
ontents of the le
ture. Expansion of logarithm into power series
will be extended to the
omplex
ase. We learn very important Abel's trans-
formation of sum. This transformation is a dis
rete analogue of integrations by
parts. Abel's theorem on the limit of power series will be applied to evaluation
of trigonometri
series related to logarithm. The
on
ept of Abel's sum of a
divergent series will be introdu
ed.
Prin
ipal bran
h of the Logarithm. Sin
e exp(x + iy) = ex(
os y + i sin y)
one gets the following formula for the logarithm: Log z = ln jz j + i Arg z , where
Arg z = arg z + 2k. We see that the logarithm is a multi-valued fun
tion, that
is why one usually
hoose a bran
h of logarithm to work. For our purposes it is
suÆ
ient to
onsider the prin
ipal bran
h of logarithm:
(3.4.1) ln z = ln jz j + i arg z; < arg z :
The prin
ipal bran
h of the logarithm is a dierentiable fun
tion of
omplex
variable with derivative 1z , inverse to exp z . This bran
h is not
ontinuous at
negative numbers. However its restri
tion on upper half-plane is
ontinuous and
even dierentiable at negative numbers.
Rz 1
Lemma 1. For any nonnegative z one has d = ln z .
1
Proof. If Im z 6= 0, the segment [0; z ℄ is
ontained in the
ir
le j z0 j < jz0 j
for z0 = Imjzj2 . In this
ir
le 1 expands into a power series, whi
h one
an
z
integrate termwise. Sin
e for z k the result of integration depends only on the
ends of integration path, the same is true for power series. Hen
e, we
an
hange
path of integration without
hanging the result. Consider R the following path:
p(t) =
os t + i sin t, t 2 [0; arg z ℄. We know the integral 1 d = i arg z . This
p
path terminates at jzzj . Continue this path by the linear path to z . The integral
Rz 1 jRzj jzj z
d = 1 dtz=jz j = R 1 dt = ln jz j. Therefore R 1 d = R 1 d +
z=jzjt t
z=jzj 1 1 1 p
z
R 1
d = i arg z + ln jz j.
z=jzj
93
In parti
ular for z = ix, one has 1+ ix 1+ix
1 ix = 1 and arg 1 ix = 2 ar
tg x. Therefore
one gets
X1 x2k+1
(3.4.4) ar
tg x = ( 1)k
k=0
2k + 1
sin = ar
tg tan(=2) = . The substitution of
Sin
e arg(1 + ei ) = ar
tg 1+
os 2
1
P ik
exp(i) for z in the Mer
ator series ln(1 + ei ) = ( 1)k+1 e k gives for the
k=1
imaginary parts:
X1 sin k
(3.4.5) ( 1)k+1 =
k=0
k 2
However the last substitution is not
orre
t, be
ause jei j = 1 and (3.4.2) is
proved only for jz j < 1. To justify it we will prove a general Theorem, due to
Abel.
Summation by parts. Consider two sequen
es fak gnk=1 , fbk gnk=1 . The dif-
feren
e of their produ
t ak bk = ak+1 bk+1 ak bk
an be presented as
(3.4.6) (ak bk ) = ak+1 bk + bk ak
Summation of these equalities gives
nX1 nX1
(3.4.7) an bn a1 b1 = ak+1 bk + bk ak
k=1 k=1
A permutation of the latter equality gives so
alled Abel's transformation of
sums
nX1 nX1
(3.4.8) bk ak = an bn a1 b1 ak+1 bk
k=1 k=1
94
Applying the Abel transformation for any m > n one gets
(3.4.10)
m
X m
X k 1
X
A(n; m) A(n; m)(x) = ak (1 xk ) = (1 x) A(n 1; k 1) xj =
k=n k=n j =0
0 1
Xm n
X m
X
(1 x) A(n 1; m) xj A(n 1; n) xj A(n 1; k)xk A
j =0 j =0 k=n
By (3.4.9) for n > N , one gets jA(n 1; m)j = j(A(0; m) A) + (A A(0; n))j
"=9 + "=9 = 2"=9. Hen
e, we
an estimate the absolute value of the expression
in the bra
kets (3.4.10) from above by 21"=x3 . As result we get
Leibniz series. As the rst appli
ation of the Abel Theorem evaluate the
1 ( 1)k
P
Leibniz series 2n+1 .
This series
onverges by the Leibniz Theorem. By the
k=0
P1 ( 1)k xk
Abel Theorem its sum is x!lim
1 0 k=0 2n+1
= x!lim
1 0
ar
tg x = ar
tg 1 = 4 . We
get the following remarkable equality:
1 1 1 1
(3.4.15) =1 + + :::
4 3 5 7 9
1
P
Abel sum of series. One denes the Abel sum of a series ak as the limit
k=0
1
P
lim a xk .
x!1 0 k=0 k
The series whi
h has Abel sum are
alled Abel summable. The
Abel Theorem shows that all
onvergent series has Abel sum
oin
iding with its
usual sum. However there are a lot of series, whi
h has Abel sum, but do not
onverge.
95
1
P
Abel's inequality. Consider a series ak bk , where partial sums An =
k=1
nP1
ak are bounded by some
onstant A and the sequen
e fbk g is monotone.
k=1
nP1 nP1 nP1 nP1
Then ak b k = bk Ak = An bn A1 b1 + Ak+1 bk . Sin
e jbk j =
k=1 k=1 k=1 k=1
jbn b1 j, one gets the following inequality:
nX1
(3.4.16)
ak bk
3A maxfjbk jg
k=1
Convergen
e test.
nP1
Theorem 3. Let the sequen
e of partial sums ak be bounded, and fbk g be
k=1
P1
non-in
reasing and innitesimally small. Then ak bk
onverges to its Abel
k=1
sum, if the latter exists.
nP1
Proof. The dieren
e between a partial sum ak bk and the Abel sum is equal
k=1
to
nX1 X 1
(3.4.17) lim a b (1 xk ) + lim ak bk xk
x!1 0 k k x!1 0
k=1 k=n
The rst limit is zero, the se
ond limit
an be estimated by Abel's inequality
from above by 3Abn. It tends to 0 as n tends to innity.
Appli
ation. Now we are ready to prove the equality (3.4.5). The series
1
P
( 1)k+1 sinkkx has an Abel sum. Indeed,
k=1
1 k 1 ix k
k+1 q sin kx = Im lim k+1 (qe ) =
X X
lim
q!1 0
( 1) q!1 0
( 1)
k=1 k k=1 k
Im q!lim
1 0
ln(1 + qeix ) = Im ln(1 + eix ):
nP1 nP1 inx
The sums sin kx = Im eikx = Im 11 eeix are bounded. And k1 is de
reas-
k=1 k=1
ing and innitesimally small. Hen
e we
an apply the Theorem 3.
Problems.
1. Evaluate 1 + 12 31 1
4 + 51 + 61 1
7
1 :::
8
P1 sin 2k
2. Evaluate k
k=1
96
1
os n
P
3. n = ln j2 sin 2 j, (0 < jj )
k=1
1 sin n
P
4. n = 2 , (0 < < 2)
k=1
1
os(2n+1)
P
5. 2n+1 = 21 ln j2
ot 2 j, (0 < jj < )
k=0
1 sin(2n+1)
P
6. 2n+1 = 4 , (0 < < )
k=0
1
P
7. ( 1)k+1
osnn = ln 2
os 2 , ( < < )
k=1
8. Find the Abel sum of 1 1 + 1 1 : : :
9. Find the Abel sum of 1 1 + 0 + 1 1 + 0 : : :
10. A periodi
series, su
h that the sum of the period is zero, has an Abel
sum.
P1 k2
11. Teles
ope
k=1 2k
nP1
12. Evaluate k
os kx
k=0
1 sin kx
P
13. Estimate from above k2
k=n
P1 P1 P1 P1
14. If ak , bk and their
onvolution
k
onverge then
k =
k=0 k=0 k=0 k=0
P1 1
P
ak bk
k=0 k=0
97
3.5 Residues Theory
On the
ontents of the le
ture. At last, the reader learns something, whi
h
Euler did not know, and what he would highly appre
iate. The residues theory
allows one to evaluate a lot of integrals whi
h were not a
essible by Newton-
Leibniz formula.
98
dierent parameterizations are obtained from ea
h other by an in
reasing
hange
of the variable (Lemma 1).
One denes a partition of a
urve by a point x as a pair of monotone
urves 1 , 2 , su
h that = 1 [ 2 and 1 \ 2 = x. And we write in this
ase = 1 + 2 .
The Partition Rule for the line integral is
Z Z Z
(3.5.1) f (z ) dg(z ) = f (z ) dg(z ) + f (z ) dg(z );
1+ 2 1 2
where the orientations on i are indu
ed by an orientation of . To prove the
Partition Rule
onsider a positive parameterization p : [a; b℄ ! . Then the
restri
tions of p over [a; p 1 (x)℄ and [p 1 (x); b℄ give a positive parameteriza-
p R1 (x)
tions of 1 and 2. Hen
e the equality (3.5.1) follows from f (z ) dg(z ) +
a
Rb Rb
f (z ) dg(z ) = f (z ) dg(z ).
p 1 (x) a
A sequen
e of oriented monotone
urves f k gnk=1 with disjoint interiors is
Pn
alled a
hain of monotone
urves and denoted by k . The body of a
hain
k=1
n
P
C= k is dened as [k=1 j k j and denoted by jC j. The interior of the
hain
n
k=1
is dened as the union of interiors of its elements. R
The integral of a form f dg along the
hain is dened as f dg =
Pn k
k=1
n R
P
f dg.
k=1 k
Pn Pm
One says that two
hains k and 0 have the same orientation, if the
k
k=1 k=1
orientations indu
ed by k and 0j on k \ 0j
oin
ide in the
ase when k \ 0j
has a nonempty interior. Two
hains with the same body and orientation are
alled equivalent.
n
P m
P
Lemma 2. If two
hains C = C0 = 0 are equivalent then the
k and
k
k=1 k=1
integrals along these
hains
oin
ide for any form fdg.
Proof. For any interior point x of the
hain C , one denes the subdivision of C
Pn
by x as +j + j + k [k 6= j ℄, where j is the
urve
ontaining x and j + j
+
k=1
is the partition of by x. The subdivision does not
hange the integral along
the
hain due to the Partition Rule.
Hen
e we
an subdivide C step by step by endpoints of C 0 to
onstru
t a
hain Q whi
h endpoints in
lude all endpoints of P 0 . And the integral along Q
is the same as along P . Another possibility to
onstru
t Q is to subdivide C 0
by endpoints of C . This
onstru
tion shows that the integral along Q
oin
ides
with the integral along C 0 . Hen
e the integrals along C and C 0
oin
ide.
Due to this lemma, one
an introdu
e the integral of a dierential form along
any oriented pie
ewise monotone
urve . To do this one
onsider a monotone
99
partition of into a sequen
e f k gnk=1 of monotone
urves with disjoint interiors
and equip all k with indu
ed orientation. One gets a
hain and the integral
along this
hain does not depends on partition.
100
On the other hand
I Z I Z
f (z )dz = f (z ) dz + L and f (z )dz = f (z ) dz L;
D0 0D D00 00 D
hen
e
I I Z Z I
f (z )dz + f (z )dz = f (z ) dz + f (z ) dz = f (z )dz
D0 D00 0D 00 D D
101
H H H
Sin
e f (z )dz + f (z )dz = f (z )dz by the Lemma 3 we get the
on-
D0 D00 Dk
R
tradi
tion with the hypothesis
pk
f (z ) dz
jRk j.
Hen
e, one of the inequal-
ities (3.5.4) holds. If the rst inequality holds we set Rk+1 = Rk0 else we set
Rk+1 = Rk00 .
After
onstru
ting the sequen
e fRk g,
onsider the point z0 belonging to
\1k=1 Rk . This point belongs to D, be
ause all its neighborhood
ontains points
of D. Consider
H the linearization f (z ) = f (z0 ) + f 0 (z0 )(z z0 ) + o(z )(z z0 ).
Sin
e (f (z0 ) + f 0 (z0 )(z z0 ))dz = 0 one gets
Dk
I I
(3.5.6)
o(z )(z z0)dz =
f (z )dz
jRk j
Dk Dk
The boundary of Dk is
ontained in the union Rk [ Rk \ D. Consider a
n
P
monotone partition D = k . Sin
e the interse
tion of Rk with a monotone
k=1
urve is a monotone
urve, one
on
ludes that D \ Rk is a union of at most n
monotone
urves. As Rk
onsists of 4 monotone
urves we get that Dk is as
a body of a
hain with at most 4 + n monotone
urves.
Denote by Pk the perimeter of Rk . And suppose that o(x) is bounded in Rk
by a
onstant ok . Then jo(x)(z z0)j Pk ok for all z 2 Rk .
Sin
e diam Dk P2k by the Estimation Lemma 4, we get the following
inequality:
I
4(4 + n)Pk ok P2k
(3.5.7)
o(z )(z z0 )dz = 2(4 + n)ok Pk2
Dk
The ratio Pk2 =jRk j is
onstant for even k. Therefore the inequalities (3.5.6)
jRk j =
jRj . However the
and (3.5.7)
ontradi
t ea
h other for ok < 2(4+ n)Pk2 2(4+n)P 2
j j
inequality jo(x)j < 2(4+n)P 2 holds for some neighborhood V of z0 as o(x) is
R
innitesimally small at z0 . This is a
ontradi
tion, be
ause V
ontains some
R2k .
Hr
Residues. By f (z ) dz we denote the integral along the boundary of the disk
z0
fjz z0 j rg.
Lemma 6. If a fun
tion f (z ) is
omplex dierentiable in the domain D with
H n Hr
P
the ex
eption of a nite set of points fzk gnk=1 . Then f (z )dz = f (z ) dz ,
D k=1 zk
where r is so small that all disks jz zk j < r are
ontained in D and disjoint.
Proof. Denote by D0 the
omplement of the union of the disks in D. Then D0 is
H union of D and the boundary
ir
les of the disks. By the Cau
hy HTheorem
the
f (z )dz = 0. On the other hand this integral is equal to the sum f (z )dz
D0 D
and the sum of integrals along boundaries of the
ir
les. The orientation indu
ed
102
by D0 onto the boundaries of these
ir
les is opposite to the orientation indu
ed
H H Pn Hr
from the
ir
les. Hen
e 0 = f (z )dz = f (z )dz f (z ) dz .
D0 D k=1 zk
A singular point of a
omplex fun
tion is dened as a point where either the
fun
tion or its derivative are not dened. A singular point is
alled isolated, if
it has a neighborhood, where it is the only singular point.
One denes the residue of f at a point z0 and denotes it as resz0 f as the
r
limit rlim 1 H f (z )dz . The above lemma shows that this limit exists for any
!0 2 i
z0
isolated singular point and moreover, that all integrals along suÆ
iently small
ir
umferen
es in this
ase are the same.
Sin
e in all regular points the residues are 0 the
on
lusion of the Lemma 6
for a fun
tion with nitely many singular points
an be presented in the form:
I X
(3.5.8) f (z )dz = 2i resz f
D z2D
103
Problems.
H1 1
1. Evaluate
1 1 + z4
H1 dz
2. Evaluate sin z
0
H1 dz
3. Evaluate ez 1
0
H1 dz
4. Evaluate z2
0
H1
5. Evaluate sin z1 dz
0
H1 1
6. Evaluate ze z dz
0
5H=2
7. Evaluate z 2
ot z dz
0
12
H z dz
8. Evaluate (z 1)(z 2)2
2
R
+
d
9. 5+3
os
+R
10. d
(1+
os2 )2
2R
11. d
(1+
os )2
0
+R1
12. dx
4
1 1+x
+R1
13. dx
(1+x2 )(4+x2 )
0
+R1
14. 1+x2
1+x4
1
+R1 3
15. x dx
6
1 1+x
104
3.6 Analyti
Fun
tions
On the
ontents of the le
ture. This le
ture introdu
es the reader into
phantasti
ally beautiful world of analyti
fun
tions. Integral Cau
hy formula,
Taylor series, Fundamental Theorem of Algebra. The reader will see all of these
treasures in a single le
ture.
Theorem 1 (Integral Cau
hy Formula). If fun
tion f is
omplex dieren-
tiable in the domain D, then for any interior point z 2 D one has
I
1 f ( ) dz
(3.6.1) f (z ) =
2i z
D
Proof. The fun
tion zf (zz)0 has its only singular point inside the
ir
le. This
is z0 , whi
h is a simple pole. The residue of zf (zz)0 by Lemma 7 is equal to
lim (z z0) zf (zz)0 = zlim
z!z0 !z0 f (z ) = f (z0 ). And by the formula (3.5.8) the integral
is equal to 2if (z0).
P1
Lemma 2. Let fk be a series of virtually monotone
omplex fun
tions,
k=1
1
P
whi
h is termwise majorized by a
onvergent positive series
k on a monotone
k=1
urve (that is jfk (z )j
k for natural k and z 2 ) and su
h that F (z ) =
P1
fk (z ) is virtually monotone. Then
k=1
1Z
X 1
Z X
(3.6.2) fk (z ) dz = fk (z ) dz
k=1 k=1
105
Proof. The proof is by
ontradi
tion. Suppose that f is unbounded. Divide the
interval [a; b℄ in half. Then the fun
tion has to be unbounded at least on one of
the halves. Consider this half and divide it in half. Choose the half where the
fun
tion is unbounded. So we
onstru
t a nested innite sequen
e of intervals
onverging to a point, whi
h
oin
ides with the interse
tion of all the intervals.
And f is obviously not lo
ally bounded at this point.
Corrolary 4. A
omplex fun
tion f (z )
ontinuous on the boundary of a domain
D is bounded on D.
Proof. Consider a path p : [a; b℄ ! D. Then jf (p(t))j is
ontinuous on [a; b℄,
hen
e it is lo
ally bounded, hen
e it is bounded. Sin
e D
an be
overed by
images of nitely many pathes this implies boundedness of f over D.
Theorem 5. If a fun
tion f (z ) is
omplex dierentiable in the disk jz z0j R,
then for jz z0 j < R
1 IR
X f ( )
(3.6.4) f (z ) = (z z0 )k d;
k=0
( z0 )k+1
z0
where the series on the right-hand side absolutely
onverges for jz z0 j < R.
Proof. Fix a point z su
h that jz z0 j < R and
onsider as a variable. For
j z0 j > jz z0 j one has
1 1 1 1 1
X (z z0 )k
(3.6.5) = = z z0 = k+1
z ( z0 ) (z z0 ) z0 1 z0 k=0 ( z0 )
On the
ir
le j z0 j = R the series on the right-hand side is majorized by a
P1 jz z0 jk
onvergent series Rk+1 for r > jz z0 j. The fun
tion f ( ) is bounded on
k=0
j z0j = R by the Corollary 4. Therefore after multipli
ation of (3.6.5) by f ( )
all
onditions of the Lemma 2 are satised. The termwise integration gives:
IR 1 IR
f ( ) X
k f ( ) d
(3.6.6) f (z ) = d = (z z0 )
z k=0 ( z0 )k+1
z0 z0
106
Theorem 7. If f is analyti
at z then f 0 is analyti
at z . If f and g are
analyti
at z then f + g, f g, fg are analyti
at z . If f is analyti
at z and g
is analyti
at f (z ) then g(f (z )) is analyti
at z .
Proof. The termwise dierentiation of a power series representing f in a neigh-
borhood of z gives the power series for its derivative. Hen
e f 0 is analyti
. The
dierentiability of f g, fg and g(f (z )) follows from
orresponding dierenti-
ation rules.
Lemma 8 (on isolated zeros). If f (z ) is analyti
and is not identi
ally equal
to 0 in some neighborhood of z0 , then f (z ) 6= 0 for all z 6= z0 suÆ
iently
lose
to z0 .
P1
Proof. Let f (z ) =
k (z z0 )k in a neighborhood U of z0 . Let
m be the rst
k=0
1
P
nonzero
oeÆ
ient. Then
k (z z0 )k m
onverges in U to a dierentiable
k=m
fun
tion g(z ) by Theorem 9. Sin
e g(z0 ) =
m 6= 0 and g(z ) is
ontinuous at
z0 , the inequality g(z ) 6= 0 holds for all z suÆ
iently
lose to z0 . As f (z ) =
g(z )(z z0 )m , the same is true for f (z ).
P1
Theorem 9 (Uniqueness Theorem). If two power series ak (z z0 )k
k=0
1
P
and bk (z z0 )k
onverge in a neighborhood of z0 and their sums
oin
ide
k=0
for some innite sequen
e fzk g1 k=1 su
h that zk 6= z0 for all k and klim z =z ,
!1 k 0
then ak = bk for all k.
P1
Proof. Set
k = ak bk . Then f (z ) =
k (z z0 )k has a non-isolated zero at
k=0
z0 . Hen
e f (z ) = 0 in a neighborhood of z0 . We get a
ontradi
tion by
onsid-
P1
ering fun
tion g(z ) =
k (z z0 )k m , whi
h is nonzero for all z suÆ
iently
k=m
lose to z0 (
f. the proof of Lemma on isolated zero), and satises equation
f (z ) = g(z )(z z0)m .
Taylor series. Set f (0) = f and by indu
tion dene the (k + 1)-th deriva-
tive f (k+1) of f as the derivative of its k-th derivative f (k) . For the rst and
the se
ond derivatives one prefers notation f 0 and f 00 . For example, the k-th
derivative of z n is nk z n k . (Re
all that nk = n(n 1) : : : (n k + 1))
The following series is
alled the Taylor series of a fun
tion f at point z0 :
X1 f (k) (z )
0
(3.6.7) (z z0 )k
k=0
k !
The Taylor series is dened for any analyti
fun
tion, be
ause an analyti
fun
tion has derivative of any order due to Theorem 7.
Theorem 10. If a fun
tion f is analyti
in the disk jz z0 j < r then f (z ) =
1 f (k) (z0 )
P
k! (z z0 )k for any z from the disk.
k=0
107
Proof. By the Theorem 5, f (z ) is presented in the disk by a
onvergent power
1
P
series ak (z z0 )k . To prove our theorem we prove that
k=0
IR
f ( ) f (k) (z0 )
(3.6.8) ak = d = :
( z0 )k+1 k!
z0
P1
Indeed, a0 = f (z0) and termwise dierentiating of ak (z z0 )k applied n
k=0
1 n
P
times gives f (n)(z ) = k ak (z z0 )k . Putting z = z0 , one gets f (n)(z0 ) =
k =n
nn an = an n!.
Theorem 11 (Liouville). If a fun
tion f is analyti
and bounded on the whole
omplex plane, then f is
onstant.
1
P
Proof. If f is analyti
on the whole plane then f (z ) = ak z k , where ak is
k=0
dened by (3.6.8). If jf (z )j B by Lemma on Estimate one gets
IR
f ( )
4 4 RBk+1 pR C
(3.6.9) jak j =
z k+1 d = k
2 R
0
Consequently ak for k > 0 is innitesimally small as R tends to innity. But ak
does not depend on R, hen
e it is 0. Therefore f (z ) = a0 .
Theorem 12 (Fundamental Theorem of Algebra). Any non
onstant poly-
nomial P (z ) has a
omplex root.
Proof. If P (z ) has no roots the fun
tion f (z ) = P (1z) is analyti
on the whole
!1 f (z ) = 0 the inequality jf (z )j < 1 holds for jz j = R if R is
plane. Sin
e zlim
suÆ
iently large. Therefore the estimation (3.6.9) for k-th
oeÆ
ient of f holds
with B = 1 for suÆ
iently large R. Hen
e the same arguments as in proof of
Liouville Theorem show that f (z ) is
onstant. This is a
ontradi
tion.
A
singular points
. .
. .
B
Analyti
ontinuation.
108
Lemma 13. If an analyti
fun
tion f (z ) has nitely many singular points in
a domain D and a non isolated zero at a point z0 2 D then f (z ) = 0 for all
regular z 2 D
Proof. For any nonsingular point z 2 D, we
onstru
t a sequen
e of suÆ
iently
small disks D0 ; D1 ; D2; : : : Dn without singular points with the following prop-
erties: 1) z0 2 D0 U , 2) z 2 Dn , 3) zk , the
enter of Dk , belongs to Dk 1 for
all k > 0. Then by indu
tion we prove that f (Dk ) = 0. First step: if z0 is a
non-isolated zero of f , then the Taylor series of f vanishes at z0 by Uniqueness
theorem. But this series represents f (z ) on D0 due to Theorem 10, sin
e D0
does not
ontain singular points. Hen
e, f (D0 ) = 0. Suppose we have proved
already that f (Dk ) = 0. Then zk+1 is a non-isolated zero of f by the third
property of the sequen
e fDk gnk=0 . Consequently, the same arguments as above
for k = 0 prove that f (Dk+1 ) = 0. And nally we get f (z ) = 0.
Consider any formula whi
h you know from s
hool about trigonometri
fun
-
tions. For example, tan(x + y) = 1tantanx+tan y
x tan y . The above Lemma implies that
this formula remains true for
omplex x and y. Indeed,
onsider the fun
tion
T (x; y) = tan(x + y) 1tantanx+tan y
x tan y . For a xed x fun
tion T (x; y ) is analyti
and
has nitely many singular points in any disk. This fun
tion has non-isolated
zeros in all real points, hen
e this fun
tion is zero in any disk interse
ting the
real line. This implies that T (x; y) is zero for all y. The same arguments applied
to T (x; y) with xed y and variable x prove that T (x; y) is zero for all
omplex
x; y.
The same arguments prove the following theorem
Theorem 14. If some analyti
relation between analyti
fun
tions holds on
a
urve , it holds for any z 2 C , whi
h
an be
onne
ted with by a paths
avoiding singular points of the fun
tions.
Lemma 15. sin t 2t for t 2 [0; =2℄
Proof. Let f (t) = sin t 2t . Then f 0 (x) =
os t 2 . Set y = ar
os 2 . Then
f 0 (x) 0 for x 2 [0; y℄. Therefore f is nonde
reasing on [0; y℄, and nonnegative,
be
ause f (0) = 0. On the interval [y; =2℄ the derivative of f is negative. Hen
e
f (x) is non-in
reasing and nonnegative, be
ause its value on the end of the
interval is 0.
Lemma 16 (Jordan). Let f (z ) be an analyti
fun
tion in the upper half-plane
!1 f (z ) = 0. Denote by R the upper half of the
ir
le jz j = R.
su
h that zlim
Then for any natural m
Z
(3.6.10) lim f (z ) exp(miz ) dz = 0
R!1
R
−r r
−R R
Consider the domain D(R) bounded by semi
ir
les (r), (R) and intervals
iz
[ R; r℄ and [r; R℄, where r = R1 and R > 1. The fun
tion ez has not singular
H eiz
points inside D(R). Hen
e z dz = 0. Hen
e for any R
D(R)
Z R iz ZR iz Z Z
e e eiz eiz
(3.6.13) dz + dz = dz dz
z z z z
r r (r) (R)
The se
ond integral on the right-hand side tends to 0 as R tends to innity due
iz
to Jordan's Lemma. The fun
tion ez has a simple pole at 0, hen
e the rst
iz
integral on the right-hand side of (3.6.13) tends to i res ez = i due to Remark
110
1. As a result, the right-hand side of (3.6.13) tends to i as R tends to innity.
Consequently the left-hand side of (3.6.13) also tends to i as R ! 1. The
RR sin x Rr sin x
imaginary part of left-hand side of (3.6.13) is equal to x dx x dx.
R r
RR
The last integral tends to 0 as r ! 0, be
ause j sinx x j 1. Hen
e sin x dx
x
R
+R1
tends to as R ! 1. Finally sin x dx = .
1 x
Problems.
1. Prove that an even analyti
fun
tion f , i.e_a fun
tion that f (z ) = f ( z )
has Taylor series at 0
onsisting only of even powers
2. Prove that analyti
fun
tion whi
h has Taylor series only with even powers
is an even fun
tion.
3. If an analyti
fun
tion f (z ) takes real values on [0; 1℄, then f (x) is real for
any real x.
+R1 1
4.
1 1 + x4
R
+ d
5.
5 + 3
os
R1 2
6. Evaluate (x2 +x a2 )2 dx (a > 0)
0
+R1
7. x sin x dx
2
1 x +4x+20
R1 ax
8. x
os
2 +b2 dx (a; b > 0)
0
111
Chapter 4
Dieren es
112
4.1 Newton Series
On the
ontents of the le
ture. The formula with binomial series was
engraved on Newton's gravestone in 1727 at Westminster Abbey.
113
P (x). Then
1
k P (0)
X X1 k P (0) X1 k P (0)
(4.1.2) Q(x) = (x + 1)k xk = xk =
k=0
k ! k=0 k ! k=0 k !
X1 k P (0) X1 k P (0) X1 k (P (0))
kxk 1 = xk 1 = xk = P (x)
k=0 k ! k=0 (k 1)! k=0 k !
Hen
e (P (x) Q(x)) = 0 and P (x) = Q(x) +
. Sin
e P (0) = Q(0), one gets
= 0. This proves P (x) = Q(x).
Lemma 4 (Lagrange Formula). For any sequen
e fyk gnk=0 , the polynomial
n
P n+1
Ln (x) = ( 1)n k k!(nyk k)! xx k has the property Ln (k) = yk for 0 k n.
k=0
n
P n+1
Proof. For x = k, all terms of the sum ( 1)n k k!(nyk k)! xx k but k-th vanish,
k
k=0
and xx k is equal to k!(n k)!( 1)n k .
Lemma 5. For any fun
tion f and for any natural number m n one has
n k f (0)
P
f (m) = k! mk .
k=0
Proof. Consider the Lagrange polynomial Ln su
h that Ln (k) = f (k) for k
n. Then k Ln (0) = k f (0) for all k n and k Ln (0) = 0 for k > n,
1 k f (0) k
P Pn k f (0) k
be
ause the degree of Ln is n. Hen
e, Ln (x) = k! x = k! x
k=0 k=0
by Lemma 3. Putting x = m in the latter equality, one gets f (m) = Ln (m) =
Pn k f (0) k
k! m .
k=0
We see that the Newton polynomial gives a solution for the interpolation
problem and our next goal is to estimate the interpolation error.
114
Theorem 7 (Rolle). If a fun
tion f is
ontinuous on the interval [a; b℄, dif-
ferentiable in interval (a; b) and f (a) = f (b), then f 0 (
) = 0 for some
2 (a; b).
Proof. If the fun
tion f is not
onstant on [a; b℄ then either its maximal or its
minimal value diers from f (a) = f (b). Hen
e at least one of its extremal values
is taken in some point
2 (a; b). Then f 0 (
) = 0 by Lemma 1.
Lemma 8. If an n-times dierentiable fun
tion f (x) has (n + 1) roots in the
interval [a; b℄, then f (n)( ) = 0 for some 2 (a; b)
Proof. The proof is by indu
tion. For n = 1 this is Rolle Theorem. Let fxk gnk=0
be a sequen
e of roots of f . By Rolle's theorem any interval (xi ; xi+1 )
ontains
a root of f 0. Hen
e f 0 has n 1 roots, and its (n 1) derivative has a root. But
the (n 1)-th derivative of f 0 is the n-th derivative of f .
Theorem 9 (Newton Interpolation Formula). Let f be an (n + 1) times
dierentiable fun
tion on I [0; n℄. Then for any x 2 I there is 2 I su
h that
n
X k f (0) k f (k+1) ( ) k+1
f (x) = x + x
k=0 k ! (k + 1)!
Proof. The formula holds for x 2 f0; 1; : : : ng and any , due to Lemma 5,
be
ause xn+1 = 0 for su
h x. For other x one has xn+1 6= 0, hen
e there is C su
h
n k f (0) k
P k+1 . The fun
tion F (y ) = f (y ) P n k f (0) k
that f (x) = k! x + Cx k! x
k=0 k=0
Cyk+1 has roots 0; 1; : : : ; n; x. Hen
e its n + 1-th derivative has a root 2 I .
n k f (0) k
P
Sin
e k! x is polynomial of degree n its (n + 1)-th derivative is 0. And
k=0
the (n +1)-th derivatives of Cxn+1 and Cxn+1
oin
ide, be
ause their dieren
e
is a polynomial of degree n. Hen
e 0 = F (n+1) ( ) = f (n+1) ( ) C (n + 1)! and
(n+1)
C = f (n+1)!() .
1 k f (0) k
P
Binomial series. The series x is
alled the Newton series of a
k!
k=0
fun
tion f . The Newton series
oin
ides with the fun
tion at all natural points.
And sometimes it
onverges to the fun
tion. The most important example of
su
h
onvergen
e is given by the so-
alled binomial series.
Consider the fun
tion (1 + x)y . This is a fun
tion of two variables. Fix x
and evaluate its dieren
e with respe
t to y. One has y (1+ x)y = (1+ x)y+1
(1 + x)y = (1 + x)y (1 + x 1) = x(1 + x)y . This simple formula allows us
immediately to evaluate ky (1 + x)y = xk (1 + x)y . Hen
e the Newton series for
(1 + x)y as fun
tion of y is
X1 xk y k
(4.1.3) (1 + x)y =
k=0 k !
For xed y and variable x, the formula (4.1.3) represents the famous Newton
binomial series. Our proof is not
orre
t. We apply Newton interpolation for-
mula, proved only for polynomials, to an exponential fun
tion. But original
Newton's proof was essentially of the same nature. Instead of interpolation of
115
the whole fun
tion, he interpolated
oeÆ
ients of its power series expansion.
Newton
onsidered the dis
overy of the binomial series as one of his greatest
dis
overies. And the role of binomial series in the further developments is very
important.
For example, Newton expands into a power series ar
sin x in the follow-
ing way. One nds the derivative of ar
sin x by dierentiating of identity
sin ar
sin x = x. This dierentiation gives
os(ar
sin x) ar
sin0 x = 1. Hen
e
ar
sin0 x =
os ar
sin
1 2 1
x = (1 x ) 2 . Sin
e
1 ( x2 )k ( 1 k 1
2) (2k 1)!! 2k
1 X X
(4.1.4) (1 x2 ) 2 = = x ;
k=0
k! k=0
2k!!
One gets the series for ar
sin by termwise integration of (4.1.4). The result
is
1
X (2k 1)!! x2k+1
(4.1.5) ar
sin x =
k=0 2k !! 2k + 1
It was more than a hundred years after the dis
overy Newton's Binomial
Theorem that it was rst
ompletely proved by Abel.
1 zk k
P
Theorem 10. For any
omplex z and su
h that jz j < 1, the series k!
k=0
absolutely
onverges to (1 + z ) = exp ( log(1 + z )).
Proof. The analyti
fun
tion exp log(1+ z ) of variable z has no singular points
in the disk jz j < 1, hen
e its Taylor series
onverges to it. The derivative of
(1 + z ) by z is (1 + z ) 1 . The k-th derivative is k (1 + z ) k . In parti
ular,
the value of k-th derivative for z = 0 is equal to k . Hen
e Taylor series of the
1 k zk
P
fun
tion is k! .
k=0
116
Plain binomial theorem. For a natural exponent the binomial series
on-
tains only nitely many nonzero terms. In this
ase it turns into (1 + x)n =
n nk xk
P n n b n
k! . Be
ause (a + b) = a (1 + a ) , one gets the following famous formula
k=0
nX
+1 k
n
(4.1.7) (a + b)n = ak bn k :
k=0
k!
This is the formula that is usually
alled Newton Binomial Theorem. But this
simple formula was known before Newton. In Europe it was proved by Pas
al
in 1654. Newton's dis
overy
on
erns the
ase of non integer exponents.
117
The binomial theorem was the main tool for expansion of fun
tions into
power series in Euler's times. Euler also applied it to get power expansions for
trigonometri
fun
tions.
Taylor expansion for x = 1 gives a symboli
equality S = exp D. Hen
e
1
P k
D = ln S = ln(1 + ) = ( 1)k+1 k . We get a formula for numeri
al
k=1
dierentiation. The symboli
al
ulations produ
e formulas whi
h hold at least
for polynomials.
Problems.
Pn nk xk yn k
1. Prove (x + y)n = k!
k=0
Pn nk
2. Evaluate n k
k! 2
k=0
k
3. If p is prime, then pk! is divisible by p.
4. nk = nn k
k! (n k)!
5. Dedu
e plain binomial theorem from multipli
ation of series for exponenta.
6. Dedu
e formula for Catalan numbers.
7. Prove that k xn xm = 0 for k < n.
n
P k
8. Prove that ( 1)k nk! = 0
k=0
9. Get a dierential equation for binomial series and solve it.
n nk k n k
P
10. Prove (a + b)n = k! a b
k=0
11. A sequen
e fak g su
h that 2 ak 0 satises inequality maxfa1; : : : an g
ak for any k between 1 and n
P1 2k
12. ( 1)k x2k! = 2x=2
os x
4
k=0
1
P 2k+1
13. ( 1)k (2xk+1)! = 2x=2 sin x
4
k=0
14. n 0p is divisible by p!
nP1
15. Prove that n 0p = ( 1)n k nk! kp
k
k=0
16. Prove
os2 x + sin2 x = 1 via power series.
118
4.2 Bernoulli numbers
On the
ontents of the le
ture. In this le
ture we give expli
it formulas for
the teles
oping of powers. These formulas involve a remarkable numbers, whi
h
were dis
overed by Ja
ob Bernoulli. They will appear in formulas
2
for sums of
series of re
ipro
al powers. In parti
ular, we will see that 6 , the sum of Euler
series,
ontains the se
ond Bernoulli number 61 .
119
Proof. Dierentiation of m+1 (x) = xm gives 0m+1 (x) = mxm 1 . The
polynomial mm has the same dieren
es, hen
e (0m+1 (x) mm (x)) = 0.
By the Lemma 2 this implies that 0m+1 (x) mm (x) is
onstant. Therefore,
0m+1 (x) mm (x) = 0m+1 (0) mm (0). But m (1) = 0 and m (0) = m (1)
m (0) = 0 0m 1 = 0.
120
P0 (x) = 1,
R1
Pn (x) dx = 0 for n > 0,
0
Pn0 (x) = nPn 1 (x) for n > 0,
then Pn (x) = Bn (x) for all n.
Analyti
properties.
Lemma 3 (on re
e
tion). Bn (x) = ( 1)nBn (1 x) for any n.
Proof. We prove that the sequen
e Tn (x) = ( 1)n Bn (1 x) satises all
ondi-
R1 R0
tions of the Theorem 2. Indeed, T0 = B0 = 1, Tn (x) dx = ( 1)n Bn (x) dx =
0 1
0 and
Tn (x)0 = ( 1)n Bn0 (1 x) = ( 1)n nBn 1 (1 x)(1 x)0 =
( 1)n+1 nBn 1 (x) = nTn 1(x)
Lemma 4 (on roots). For any odd n > 1 the polynomial Bn (x) has on [0; 1℄
just three roots: 0; 12 ; 1.
Proof. For odd n, the re
e
tion Lemma 3 implies that Bn ( 12 ) = Bn ( 21 ), that
is Bn ( 12 ) = 0. Furthermore, for n > 1 one has Bn (1) Bn (0) = n0n 1 =
0. Hen
e Bn (1) = Bn (0) for any Bernoulli polynomial of degree n > 1. By
re
e
tion formula for an odd n one obtains Bn (0) = Bn (1). Thus any Bernoulli
polynomial of odd degree greater than 1 has roots 0; 12 ; 1.
The proof that there are no more roots is by
ontradi
tion. In the opposite
ase
onsider Bn (x), of the least odd degree > 1 whi
h has a root dierent
from the above mentioned numbers. Say < 21 . By Rolle's Theorem Bn0 (x)
has at least three roots 1 < 2 < 3 in (0; 1). To be pre
ise, 1 2 (0; ),
2 2 (; 12 ), 3 2 ( 12 ; 1). Then Bn 1 (x) has the same roots. By Rolle's Theorem
Bn0 1 (x) has at least two roots in (0; 1). Then at least one of them diers from
1 and is a root of B
2 n 2 (x). By the minimality of n one
on
ludes n 2 = 1.
However, B1 (x) has the only root 21 . This is a
ontradi
tion.
Theorem 5. Bn = 0 for any odd n > 1. For n = 2k, the sign of Bn is ( 1)k+1 .
For any even n one has either Bn = maxx2[0;1℄ Bn (x) or Bn = minx2[0;1℄ Bn (x).
The rst o
urs for positive Bn , the se
ond for negative.
Proof. B2k+1 = B2k+1 (0) = 0 for k > 0 by the Lemma 4. Above we have
found that B2 = 61 . Suppose we have established that B2k > 0 and that this
is the maximal value for B2k (x) on [0; 1℄. Let us prove that B2k+2 < 0 and is
the minimal value for B2k+2 (x) on [0; 1℄. The derivative of B2k+1 in this
ase
is positive at the ends of [0; 1℄, hen
e B2k+1 (x) is positive for 0 < x < 12 and
negative for 12 < x < 1, by Lemma 4 on roots and Theorem on Intermediate
values. Hen
e, B20 k+2 (x) > 0 for x < 21 and B20 k+2 (x) < 0 for x > 21 . Therefore,
B2k+2 (x) takes the maximal value in the middle of [0; 1℄ and takes the minimal
121
values at the ends of [0; 1℄. Sin
e the integral of the polynomial along [0; 1℄ is
zero 0 and the polynomial is not
onstant, its minimal value has to be negative.
The same arguments prove that if B2k is negative and minimal, then B2k+2 is
positive and maximal.
Lemma 6 (Lagrange Formula). If f is a dierentiable fun
tion on [a; b℄,
then there is a 2 (a; b), su
h that
f (b) f (a)
(4.2.10) f (b) = f (a) + f 0 ( )
b a
Proof. The fun
tion g(x) = f (x) (x a) f (bb) af (a) is dierentiable on [a; b℄
and g(b) = g(a) = 0. By Rolle's Theorem g0( ) = 0 for some 2 [a; b℄.
Hen
e f 0 ( ) = f (bb) fa(a) . Substitution of this value of f 0 ( ) in (4.2.10) gives
the equality.
Generating fun
tion. The following fun
tion of two variables is
alled the
generating fun
tion of Bernoulli polynomials
X1 tk
(4.2.11) B (x; t) = Bk (x)
k=0
k!
Sin
e Bk 2kk! , the series on the right-hand side
onverges for t < 2 for any
x. Let us dierentiate it termwise as a fun
tion of x, for a xed t. We get
1
P 0 x;t)
kBk 1 (x) tk! = tB (x; t). Consequently (log B (x; t))0x = BBx((x;t
k
) = t and
k=0
log B (x; t) = xt +
(t), where the
onstant
(t) depends on t. It follows that
B (x; t) = exp(xt)k(t), where k(t) = exp(
(t)). For x = 0 we get B (0; t) =
1 tk
P
k(t) = Bk k! . To nd k(t)
onsider the dieren
e B (x + 1; t) B (x; t).
k=0
It is equal to exp(xt + t)k(t) exp(xt). On the other hand the dieren
e is
1
P k P1 k
Bk (x) tk! = kBk 1 (x) tk! = tB (x; t). Comparing these expressions we
k=0 k=0
get expli
it formulas for the generating fun
tions of Bernoulli numbers:
t X1 B
k k
(4.2.12) k(t) = = t ;
exp t 1 k=0 k!
and Bernoulli polynomials:
0 1
X tk t exp(tx)
(4.2.13) B (x; t) = Bk (x) =
k=+ k! exp t 1
1
P k
From (4.2.11) one gets t = (exp t 1) Bk tk! . Substituting exp t 1=
k=0
1 tk
P
in this equality, by the Uniqueness Theorem, one gets the equalities for
k!
k=1
the
oeÆ
ients of the power series
n
X Bn k
(4.2.14) =0 for n > 1
k=1
(n k)!k!
122
Add Bnn! to both sides of this equality and multiply both sides by n! to get
n
X Bk nk
(4.2.15) Bn = for n > 1
k=0
k!
The latter equality one memorizes via the formula B n = (B + 1)n , where after
expansion of the right hand side, one should move down all the exponents at B
turning the powers of B into Bernoulli numbers.
Now we are ready to prove that
m
B (x + 1) Bm X (m 1)k 1 k
(4.2.16) m (1 + x) = m = Bm k x = m (x)
m m k=0 k!
Putting x = 0 in the right hand side one gets m (0) = Bm (m 1) 1 = Bmm . The
left-hand side takes the same value at x = 0, be
ause Bm (1) = Bm (0) = Bm . It
remains to prove the equality of the
oeÆ
ients in (4.2.16) for positive degrees.
m k m k k j j
Bm (x + 1) 1 X m Bm k 1X m Bm k X kx
= (1 + x)k = =
m m k=0 k! m k=0 k! j=0 j !
Now let us
hange the summation order and
hange the summation index of the
interior sum by i = m k:
m jm
1X m jX
x m mk Bm k j 1 X x Xj mm i Bi
= k = (m i)j =
m j=0 j ! k=j k! m j=0 j ! i=0 (m i)!
m i j i j
Now we
hange m (m(mi)! i) by (m ij!) m and apply the identity (4.2.15)
m j jm
X x m Xj Bi (m j )i X m
(m 1)j 1 xj
= = Bm j
j =0 mj ! i=0 i! j =0 j!
Problems.
R1
1. Evaluate Bn (x) sin 2x.
0
2. Expan dx 4 3x2 + 2x 1 as a polynomial in (x 2).
3. Cal
ulate the rst 20 Bernoulli numbers.
4. Prove the inequality jBn (x)j jBn j for even n.
5. Prove the inequality jBn (x)j n4 jBn 1j for odd n.
R1 R1
6. Prove that f (0)+2 f (1) = f (x) dx + f 0 (x)B1 (x) dx
0 0
R1 0 R1
7. Prove that f (0)+2 f (1) = f (x) dx + f2(0) f 00 (x)B2 (x) dx
0 0
123
8. Dedu
e Bn (x) = nxn 1 from the balan
e property and the dierentia-
tion rule.
9. Prove that Bn (x) = Bn (1 x), using the generating fun
tion.
10. Prove that B2n+1 = 0, using the generating fun
tion.
n 1
11. Prove that Bm (nx) = nm 1 P B x + nk
m
k=0
12. Evaluate Bn ( 21 ).
13. Prove that B2k (x) = P (B2 (x)), where P (x) is a polynomial with positive
oeÆ
ient (Ja
obi Theorem).
1
P k n
14. Prove that Bn = ( 1)k k+10
k=0
P 1
15. Prove that Bm + k+1 [k + 1 is prime and k is divisor of m℄ is an inte-
ger. (Staudt theorem).
124
4.3 Euler-Ma
laurin Formula
On the
ontents of the le
ture. From this le
ture
1 1 we will learn how Euler
P
managed to
al
ulate 18 digit pla
es of the sum k2 .
k=0
125
B1 B2 B4 B6 B8 B10 B12 B14 B16 B18 B20
1 1 1 1 1 5 691 7 3617 43867 174611
2 6 30 42 30 66 2730 6 510 798 330
We see from the table that in
reasing of the number of
onsidered terms
does not improve a
ura
y noti
eably.
Proof. Sin
e Bm fxgdx = d Bmm+1+1fxg and Bm+1 fkg = Bm+1 for any natural k,
the formula (3) is obtained by a simple integration by parts.
Theorem 4. For any fun
tion f and natural numbers n and m one has:
Zn m
X1 Bk+1 (k)
(4.3.6) n1 f (k) = f (x) dx + f (n) f (k) (1)
k=1 (k + 1)!
1
Zn
( 1)m+1
+ f (m)(x)Bm fxg dx
m!
1
126
Proof. The proof is by indu
tion on m. For m = 1, formula (4.3.6) is just given
by Lemma 2. Suppose (4.3.6) is proved for m. The remainder
Zn
( 1)m+1
f (m)(x)Bm fxg dx
m!
1
an be transformed by virtue of Lemma 3 into
Zn
( 1)m+1 Bm+1 (m) ( 1)m+2
(f (n) f (m)(1)) + Bm+1 fxgf (m+1)(x) dx:
(m + 1)! (m + 1)!
1
Sin
e odd Bernoulli numbers vanish, ( 1)m+1 Bm+1 = Bm+1 for m > 0.
Z1 Z0:5 Z1
(4.3.8) f (x)B2m+1 (x) dx = f (x)B2m+1 (x) dx + f (x)B2m+1 (x) dx =
0 0 0:5
Z0:5
(f (x) f (1 x))B2m+1 (x) dx
0
B2m+1 (x) is equal to 0 at the end-points of [0; 0:5℄ and has
onstant sign on
(0; 0:5), hen
e its sign on the interval
oin
ides with the sign of its derivative
at 0, that is, it is equal to sgn B2m . The dieren
e f (x) f (1 x) also has
onstant sign as x < 1 x on (0; 0:5) and its sign is sgn(f (1) f (0)). Hen
e
the integrand in (4.3.8) has
onstant sign. Consequently the integral itself has
the same sign as the integrand has.
127
Lemma 7. If f (2m+1)(x) and f (2m+3) (x) are
omonotone for x 1 then
B2m+2 (2m+1)
(4.3.9) R2m = m (f (n) f (2m+1) (1)) 0 m 1
(2m + 2)!
Proof. The signs of R2m+1 and R2m+3 are opposite. Indeed, by Lemma 5
sgn B2m = sgn B2m+2 and sgn(f (2m+1) (n) f (2m+1)(1)) = sgn(f (2m+3) (n)
f (2m+3) (1), owing to
omonotonity
ondition. Hen
e sgn R2m+1 = sgn R2m+3
by Lemma 6.
Set T2m+2 = (2Bm2m+2)!
+2 (f (2m+1) (n) f (2m+1) (1)). Then T
2m+2 = R2m+1
R2m+2 . By Lemma 5, T2m+2 = R2m+1 R2m+3 . Sin
e R2m+3 and R2m+1 have
opposite signs, it follows that sgn T2m+2 = sgn R2m+1 and jT2m+2 j jR2m+1 j.
Hen
e m = RT22mm+2
+1 = R2m belongs to [0; 1℄.
T2m+2
Theorem 8. If f (k) and f (k+2) are
omonotone for any k > 1, then
Zn m
n X B2k (2k
(4.3.10) f (k )
1 f (x) dx f 1) (n) f (2k 1) (1)
1 k=1 (2k )!
B2m+2 (2m+1)
(2m + 2)! f
(n) f (2m+1)
Hen
e the value of the error whi
h gives the summation formula (4.3.4) with
m terms has the same sign as the rst reje
ted term, and its absolute value does
not ex
eed the absolute value of the term.
R1
Theorem 9. Suppose that jf (k) (x)j dx < 1, xlim !1 f (k) (x) = 0 and f (k) is
1
omonotone with f (k+2) for all k K for some K . Then there is a
onstant C
su
h that for any m > K for some m 2 [0; 1℄ holds
(4.3.11)
n Zn m
X f (n) X B2k (2k 1) B
f (k) = C + + f (x) dx+ f (n)+m 2m+2 f (2m+1) (n)
k=1 2
1 k=1 (2k )! (2m + 2)!
128
equality 4.3.5 to the interval [n; 1), one gets
Z1
( 1)k+1 Bk ( 1)k+1 Bk+1 (k)
(4.3.13) Bk fxgf (k) (x) dx = f (n)
k! (k + 1)!
n
Z1
( 1)k+2 Bk+1
Bk+1 fxgf (k+1) (x) dx
(k + 1)!
n
Iterating this formula one gets
(4.3.14)
Z1 m Z1
X Bk+1 (k) ( 1)m
RK = BK fxgf (K ) dx + f (n) + Bm fxgf (m)(x) dx
k=K (k + 1)! m !
1 n
Here we take into a
ount the equalities ( 1)k Bk = Bk and ( 1)m+2 = ( 1)m .
Now we substitute this expression of RK into (4.3.6) and set
Z1
f (1) KX1 Bk+1 (k)
(4.3.15) C = ( 1) K +1 BK fxgf (K )(x) dx f (1):
1
2 k=1 (k + 1)!
129
4.4 Gamma fun
tion.
On the
ontents of the le
ture. Euler's Gamma-fun
tion is the fun
tion
responsible for innite produ
ts. An innite produ
t whose terms are values of
a rational fun
tion at integers are expressed in terms of Gamma-fun
tion. In
parti
ular it will help us to prove Euler's fa
torization of sin.
Teles
oping problem. Given a fun
tion f (x), nd a fun
tion F (x) su
h that
F = f . This is the teles
oping problem for fun
tions. In parti
ular, for f = 0
any periodi
fun
tion of period 1 is a solution. In the general
ase, to any
solution of the problem we
an add a 1-periodi
fun
tion and get other solution.
The general solution has the form F (x)+ k(t) where F (x) is a parti
ular solution
and k(t) is a 1-periodi
fun
tion,
alled periodi
onstant.
The Euler-Ma
laurin formula gives a formal solution of the problem, but the
Euler-Ma
laurin series rarely
onverges. Another formal solution is
X1
(4.4.1) F (x) = f (x + k)
k=0
Trigamma. Now let us try to teles
ope the Euler series. The series (4.4.1)
onverges for f (x) = x1m provided m 2 and x 6= n for natural n > 1. In
parti
ular, the fun
tion
X1 1
(4.4.2) =(x) =
k=1
(x + k )2
is analyti
; it is
alled the trigamma fun
tion and teles
opes 1+1x2 . Its value
=(0) is just the sum of Euler series.
This fun
tion is distinguished among others fun
tions teles
oping 1+1x2 by
nite variation.
Theorem 1. There is a unique fun
tion =(x) su
h that =(x) = 1
1+x2 ,
1 1
P
var=[0; 1℄ < 1 and =(0) = k2 .
k=1
1
P 1
Proof. Sin
e = is monotone, one has var= [0; 1℄ = j=j = P k12 < 1.
k=0 k=1
Suppose f (x) is another fun
tion of nite variation teles
oping 1+1x2 . Then
f (x) =(x) is a periodi
fun
tion of nite variation. It is obvious that su
h
fun
tion is
onstant, and this
onstant is 0 if f (1) = =(1).
1
P 1
Digamma. The series x+k , whi
h formally teles
opes x1 , is divergent.
P1 k=0
k [k 6= 0℄ is
onvergent and it teles
opes
However the series 1 1 1
x+k x,
k=0
be
ause adding a
onstant does not ae
t the dieren
es. Indeed,
(4.4.3)
X1 1 1
X 1 1 1
X1 1 1
[k 6= 0℄ + [k 6= 0℄ = = :
k=0 x + 1 + k k k=0 x + k k k=0 x + k x
130
The fun
tion
1
X
1 1
(4.4.4) (x) =
+
k=1 k x+k
is
alled the digamma fun
tion. Here
is the Euler
onstant. Digamma is
an analyti
fun
tion, whose derivative is the trigamma fun
tion, and whose
dieren
e is 1+1 x .
Monotoni
ity distinguish among others fun
tion teles
oping 1+1 x .
Theorem 2. There is a unique monotone fun
tion (x) su
h that (x) = 1+1 x
and (0) =
.
Proof. Suppose f (x) is a monotone fun
tion teles
oping 1+1 x . Denote by v the
variation of f on [0; 1℄. Then the variation of f over [1; n℄ is nv. On the
Pn
other hand, varf [1; n℄ = 1
k < ln n +
. Hen
e the variation of f (x) (x)
k=1
on [1; n℄ is less than 2(
+ ln n). Hen
e v for any n satises the inequality
nv 2(
+ ln n). Sin
e nlim ln n
!1 n = 0, we get v = 0. Hen
e f is
onstant,
and it is zero if f (1) = (1).
Lemma 3. 0 = =.
R x
Proof. To prove that 0 (x) = =(x),
onsider F (x) = =(t) dt. This fun
tion
1
is monotone, be
ause F 0 (x) = =(x) 0. Further (F )0 = F 0 = =(x) =
1 1
(1+x)2 . It follows F = 1+x +
, where
is a
onstant. By the Theorem
2 it follows F (x + 1)
x
= (x). Hen
e (x)0 = F 0 (x + 1) +
= =(x).
This proves that 0 is dierentiable and has nite variation. As (x) = 1+1 x
it follows 0 (x) = (1+1x)2 . We get that 0 (x) = =(x) by theorem 1.
Teles
oping the logarithm. To teles
ope logarithm, we start with the for-
1
P P1
mal solution ln(x + k). To de
rease the divergen
e, add ln k termwise.
k=0 k=1
P1 1
P
We get ln x (ln(x + k) ln k) = ln x ln(1 + xk ). We know
k=1 k=1
that ln(1 + x) is
lose to x, but the series still diverges. Now
onvergen
e
an be rea
hed by subtra
ting of kx from the k-th term of the series. This
substra
tion
hanges the dieren
e. Let us evaluate the dieren
e of F (x) =
1
P
ln x (ln(1 + xk ) xk ). The dieren
e of the n-th term of the series is
k=1
x+1
x+1 x x
ln 1 + ln 1 + =
k
k k k
x+1 x 1
ln(x + k + 1) ln k ln(x + k) ln k = ln(x + k)
k k k
131
Hen
e
1
X
1
(4.4.5) F (x) = ln x ln(x + k) =
k=1 k
nX1 !
1
n!1
lim ln x ln(x + k) =
k=1 k
nX1 !
1
lim
n!1
ln x ln(n + x) + =
k=1
k
nX1 !
1
ln x + nlim
!1(ln(n) ln(n + x)) + nlim
!1 k=1 k ln n = ln x +
As a result, we get the following formula for a fun
tion, whi
h teles
opes the
logarithm:
X1 x x
(4.4.6) (x) =
x ln x ln 1 +
k=1
k k
Theorem 4. The series (4.4.6)
onverges absolutely for all x ex
ept negative
integers. It presents a fun
tion (x) su
h that (1) = 0 and (x) = ln x.
Proof. The inequality 1+x x ln(1 + x) x implies
2
j ln(1 + x) xj 1 +x x x = 1 x+ x :
(4.4.7)
Denote by " the distan
e from x to the
losest negative integer. Then due to
P1 y
(4.4.7), the series ln 1 + ky k is termwise majorized by the
onvergent
k=1
P1 x2
series "k2 . This proves the absolute
onvergen
e of (4.4.6).
k=1
nP1 nP1
Sin
e nlim
!1 (ln(1 + k1 ) 1
!1(ln n
k ) = nlim
1
k) =
, one gets (1) =
k=1 k=1
0.
Convexity. There are a lot of fun
tions that teles
ope the logarithm. The
property whi
h distinguishes among others is
onvexity.
Throughout the le
ture and are nonnegative and
omplementary to ea
h
other, that is + = 1. The fun
tion f is
alled
onvex if, for any x, y, it
satises the inequality:
(4.4.8) f (x + y) f (x) + f (y) 8 2 [0; 1℄
Immediately from the denition it follows that
Lemma 5. Any linear fun
tion ax + b is
onvex.
Lemma 6. Any sum (even innite) of
onvex fun
tions is a
onvex fun
tion.
The produ
t of a
onvex fun
tion by a positive
onstant is a
onvex fun
tion.
132
Lemma 7. If f (p) = f (q) = 0 and f is
onvex, then f (x) 0 for all x 2= [p; q℄.
Proof. If x > q then q = x + p for = xq pp . Hen
e f (q) f (x) + f (p) =
f (x), it follows f (x) f (q) = 0. For x < p one has p = x + q for = qq xp .
Hen
e 0 = f (p) f (x) + f (q) = f (x).
Lemma 8. If f 00 is nonnegative then f is
onvex.
Proof. Consider the fun
tion F (t) = f (l(t)), where l(t) = x + y. Newton's
formula for F (t) with nodes 0, 1 gives F (t) = F (0) + F (0)t + 12 F 00 ( )t2 .
Sin
e F 00 ( ) = (y x)2 f 00 ( ) > 0, and t2 = t(1 t) < 0 we get inequality
F (t) F (0) + tF (1). Sin
e F () = f (x + y) this is just the inequality of
onvexity.
Lemma 9. If f is
onvex, then 0 f (a) + f (a) f (a + ) 2 f (a 1) for
any a and any 2 [0; 1℄
Proof. Sin
e a + = a + (a + 1) we get f (a + ) f (a) + f (a + 1) =
f (a) + f (a). On the other hand, the
onvex fun
tion f (a + x) f (a)
xf (a 1) has roots 1 and 0. By Lemma 7 it is nonnegative for x > 0. Hen
e
f (a + ) f (a) + f (a 1). It follows that f (a) + f (a) f (a + )
f (a) + f (a) f (a) f (a 1) = 2 f (a 1)
Theorem 10. (x) is the unique
onvex fun
tion that teles
opes ln x and
(1) = 1.
Proof. Convexity of follows from the
onvexity of the summands of its series.
The summands are
onvex be
ause their se
ond derivatives are nonnegative.
Suppose there is another
onvex fun
tion f (x) whi
h teles
opes the logarithm
too. Then (x) = f (x) (x) is a periodi
fun
tion, = 0. Let us prove that
(x) is
onvex. Consider a pair
; d, su
h that j
dj 1. Sin
e f (
+ d)
f (
) f (d) 0, as f is
onvex, one has
(4.4.9) (
+ d) (
) (d) = (f (
+ d) f (
) f (d))
((
+ d) (
) (d)) (
) + (d) (
+ d):
First, prove that satises the following "-relaxed inequality of
onvexity:
(4.4.10) (
+ d) (
) + (d) + ":
In
reasing
and d by 1, we do not
hange the inequality as = 0. Due to this
fa
t, we
an in
rease
and d to satisfy
1 1 < 3" . Set L(x) = (
) + (x
) ln
.
By Lemma 9 for x 2 [
;
+1℄ one has jx L(x)j 2 (
1) = ln
ln(
1) =
ln(1 +
1 1 )
1 1 < 3" . Sin
e j(x) L(x)j < 3" for x =
; d;
+2 d , it follows that
(
) + (d) (
+ d) diers from L(
) + L(d) L(
+ d) = 0 less
than by ". The inequality (4.4.10) is proved. Passing to the limit as " tends to
0, one eliminates ".
Hen
e (x) is
onvex on any interval of length 1 and has period 1. Then
(x) is
onstant. Indeed,
onsider a pair a; b with
ondition b 1 < a < b. Then
a = (b 1) + b for = b a. Hen
e f (a) f (b) + f (b 1) = f (b).
Lemma 11. 00 (1 + x) = =(x)
133
Rx
Proof. The fun
tion F (x) = (t) dt is
onvex be
ause its se
ond derivative
1
is =. The dieren
e of F 0 = is 1+1 x . Hen
e F (x) = ln(x + 1) +
, where
is some
onstant. It follows F (x 1)
x +
= (x). Hen
e is twi
e
dierentiable and its se
ond derivative is =.
Gamma fun
tion. Now we dene Euler's gamma fun
tion (x) as exp((x)),
where (x) is the fun
tion teles
oping the logarithm. Exponentiating (4.4.6)
gives a representation of Gamma fun
tion in so-
alled
anoni
al Weierstrass
form :
e
x Y1 x 1 x
(4.4.11) (x) = 1+ ek
x k=1 k
Sin
e ln (x) = ln x, one gets the following
hara
teristi
equation of the
Gamma fun
tion
(4.4.12) (x + 1) = x (x)
Sin
e (1) = 0, a
ording to (4), one proves by indu
tion that (n) = (n 1)!
using (4.4.12).
A nonnegative fun
tion f is
alled logarithmi
ally
onvex if log f (x) is
on-
vex.
Theorem 12 (
hara
terization). (x) is the unique logarithmi
ally
onvex
fun
tion dened for x > 0, whi
h satises equation (4.4.12) for all x > 0 and
takes the value 1 at 1.
Proof. Logarithmi
al
onvexity of (x) follows from the
onvexity of (x). Fur-
ther (1) = exp (1) = 1. If f is a logarithmi
ally
onvex fun
tion satisfying
the gamma-equation, then ln f satises all
onditions of Theorem 4. Hen
e,
ln f (x) = (x) and f (x) = (x).
R1
Theorem 13 (Euler). For any x 0 one has (x) = tx 1 e t dt
0
Let us
he
k that the integral satises all
onditions of Theorem 12. For
R1
x = 1 the integral gives e t dt = e t j1 0 = 1. The integration by parts
0
R1 x t R1 x R1 t x 1
t e dt = t de t = txe t j1 0 + e xt dx proves that it satises
0 0 0
the gamma-equation (4.4.12). It remains to prove logarithmi
onvexity of the
integral.
Lemma 14 (mean
riterium). If f is a monotone fun
tion whi
h satises the
following mean inequality 2f ( x+2 y ) f (x) + f (y) for all x; y then f is
onvex.
Proof. We have to prove the inequality f (x + y) f (x) + f (y) = L()
for all , x and y. Set F (t) = f (x + (y x)t); than F also satises the mean
inequality. And to prove our lemma it is suÆ
ient to prove that F (t) L(t) for
all t 2 [0; 1℄.
First we prove this inequality only for all binary rational numbers t, that is
for numbers of the type 2mn , m 2n . The proof is by indu
tion on n, the degree
134
of the denominator. If n = 0, the statement is true. Suppose the inequality
F (t) L(t) is already proved for fra
tions with denominators of degree n.
Consider r = 2nm+1 , with odd m = 2k + 1. Set r = 2kn , r+ = k2+1 n . By the
+ +r
indu
tion hypothesis F (r ) L(r ). Sin
e r = 2 , by mean inequality one
r
+ + +
has F (r) f (r )+2 f (r ) L(r )+2 L(r ) = L( r +2 r ) = L(r).
Thus our inequality is proved for all binary rational t. Suppose F (t) > L(t)
for some t. Consider two binary rational numbers p, q su
h that t 2 [p; q℄ and
jq pj < jfF((yt)) fL((xt))j . In this
ase jL(p) L(t)j jp tjjf (y) f (x)j < jF (t) L(t)j.
Therefore F (p) L(p) < F (t). The same arguments give F (q) < F (t). This is
a
ontradi
tion, be
ause t is between p and q and its image under a monotone
mapping has to be between images of p and q.
Lemma 15 (Cau
hy-Bunyakovski-S
hwarz).
0 b 12
Z Zb Zb
(4.4.13) f (x)g(x) dxA f 2 (x) dx g2 (x) dx
a a a
Rb
Proof. Sin
e (f (x) + tg(x))2 dx 0 for all t, the dis
riminant of the following
a
quadrati
equation is non-negative:
Zb Zb Zb
(4.4.14) t2 g2 (x) dx + 2t f (x)g(x) dx + f 2 (x) dx = 0
a a a
!2
Rb Rb Rb
This dis
riminant is 4 f (x)g(x) dx 4 f 2 (x) dx g2 (x) dx.
a a a
Now we are ready to prove the logarithmi
onvexity of Euler integral. The
integral is obviously an in
reasing fun
tion, hen
e by the mean
riterion it is
suÆ
ient to prove the following inequality:
0 12
Z1 Z1 Z1
x+y
(4.4.15) t 2 1 e t dtA tx 1e t dt ty 1 e t dt
0 0 0
This inequality turns into the Cau
hy-Bunyakovski-S
hwarz inequality (4.4.13)
x 1 y 1
for f (x) = t 2 e t=2 and g(t) = t 2 e t=2
135
1
Q 1 1 1 1 1
by n2 transforms it into (1 2n ) (1 + 2n ) . Introdu
ing multipliers e 2n
k=1
1
and e 2n , one gets a
anoni
al form
Y1 1
1
1 1
Y 1
1
1
(4.4.17) 1 e 2 n 1+ e 2 n
k=1
2n k=1
2n
Problems.
1
Q 3x
1. Evaluate the produ
t 1 + nx 1 + 2nx 1 n
k=1
1
Q k(5+k)
2. Evaluate the produ
t (3+k)(2+k)
k=1
3. The sum of logarithmi
ally
onvex fun
tions is logarithmi
ally
onvex.
n!n x x n
4. Prove (x) = nlim
!1 x
1
Q k k+1 x
5. Prove x+k k = (x + 1)
k=1
6. Prove Legendre's doubling formula (2x) (0:5) = 22x 1 (x + 0:5) (x)
136
4.5 The Cotangent
On the
ontents of the le
ture. In this le
ture we perform what was
promised at the beginning: we sum up the Euler series and expand sin x into
the produ
t. We will see that sums of series of re
ipro
al powers are expressed
via Bernoulli numbers. And we will see that the fun
tion responsible for the
summation of the series is the
otangent.
1 1
P
An ingenious idea, whi
h led Euler to nding the sum k2 is the following.
k=1
One
an
onsider sin x as a polynomial of innite degree. This polynomial has as
Q of the type k . Any ordinary polynomial
an be expanded into
roots all points
the produ
t (x xk ) where xk are its roots. By analogy, Euler
onje
tured
that sin x
an be expanded into the produ
t
Y1
(4.5.1) sin x = (x k):
k= 1
This produ
t diverges, but
an be modied to a
onvergent one by division of
the n-th term by n. The division does not
hange the roots. The modied
produ
t is
Y1 x Y1 x2
(4.5.2) 1 =x 1 :
k= 1
k k=1 k2 2
Two polynomials with the same roots
an dier by a multipli
ative
onstant.
To nd the
onstant,
onsider x = 2 . In this
ase we get the inverse to Wallis
produ
t in (4.5.2) multiplied by x = 2 . Hen
e the value of (4.5.2) is 1, whi
h
oin
ides with sin 2 . Thus it is natural to expe
t that sin x
oin
ides with the
produ
t (4.5.2).
1
Q
There is another way to tame (x k). Taking logarithm, we get a
k= 1
1
P
divergent series ln(x k), but a
hieve
onvergen
e by term-wise dier-
k= 1
entiation. Sin
e the derivative of ln sin x is
ot x, it is natural to expe
t that
ot x
oin
ides with the following fun
tion
X1 1 1 X 1 2x
(4.5.3)
tg(x) = = +
k= 1 x k x k=1 x k2 2
2
z P1 B k
Cotangent expansion. The expansion =
ez 1 k! z allows us to get a
k
k=0
power expansion for
ot z . Indeed, representing
ot z by Euler's formula one
gets
eiz + e iz e2iz + 1 2i 1 2iz 1X 1 B
k
i iz = i = i + = i + = i + (2iz )k
e e iz e2iz 1 e2iz 1 z e2iz 1 z k=0 k!
The term of the last series
orresponding to k = 1 is 2izB1 = iz . Multiplied
by 1z , it turns into i, whi
h eliminates the rst i. The summand
orresponding
137
to k = 0 is 1. Taking into a
ount that B2k+1 = 0 for k > 0, we get
1 X 1 4k B2k 2k 1
(4.5.4)
ot z = + ( 1)k z
z k=1 (2k)!
138
Let us denote by Z the set fk j k 2 Zg of -integers.
Lemma 4. The set of singular points of
ot z is Z. All these points are simple
poles with residue 1.
Proof. The singular points of
ot z
oin
ide with the roots of sin z . The roots
of sin z are roots of equation eiz = e iz whi
h is equivalent to e2iz = 1. Sin
e
je2iz j = je 2 Im z j one gets Im z = 0. Hen
e sin z has no roots beyond the real
line. And all its real roots as we know have form fkg. Sin
e zlim !0 z
ot z =
z
os z z 1
lim
z!0 sin z !0 sin z = sin0 0 = 1. We get that 0 is a simple pole of
ot z
= zlim
with residue 1 and others poles have the same residue be
ause of periodi
ity of
ot z .
Lemma 5. Let f (z ) be an analyti
fun
tion on a domain D. Suppose that f
has in D nitely many singular points, they are not -integers and D has no
-integer point on its boundary. Then
(4.5.9)
I X1 X
f ( )
ot d = 2i f (k)[k 2 D℄ + 2i resz (f (z )
ot z )[z 2= Z℄
D k = 1 z 2 D
.
Proof. In our situation every singular points of f (z )
ot z in D are either -
integer or singular point of f (z ). Sin
e resz=k
ot z = 1, it follows that
resz=k f (z )
ot z = f (k). Hen
e the
on
lusion of the lemma is a dire
t
on-
sequen
e of the Residue Theory.
Exploring
tg(z ).
Lemma 6.
tg(z + ) =
tg(z ) for any z
n
P nP1
Proof.
tg(z + ) = nlim 1 = nlim 1 = nlim 1
!1 k= n
z+ k !1 k= n 1
z+k !1 z (n+1) +
(nP1)
lim 1 + lim
n!1 z n n!1 k= (n
1
z+ k = 0 + 0 +
tg(z )
1)
Lemma 7. The series representing
tg(z )
onverges for any z whi
h is not a
-integer. j
tg(z )j 2 for all z su
h that j Im z j >
Proof. For any z one has jz 2 k2 2 j k2 for k > jz j. This provides the
onvergen
e of the series. Sin
e
tg(z ) has period , it is suÆ
ient to prove
the inequality of Lemma in the
ase x 2 [0; ℄, where z = x + iy. In this
ase
jyj jxj and Re z 2 = x2 y2 0. Then Re(z 2 k2 2 ) k22 . 1It follows
jz 2 k2 2 j k2 2 . Hen
e j
tg(z )j is term-wise majorized by 1 + P k212 <
k=1
2.
Lemma 8. j
tg(z )j 3 for any z with Re z = 2 .
2
Proof. In this
ase Re(z 2 k2 2 ) = 4 y2 k2 2 k2 for all k 1. Hen
e
1
jC (z )j 2 + P k12 1 + 2 = 3.
k=1
139
Lemma 9. For any z 6= k and domain D whi
h
ontains z and whose boundary
does not
ontain -integers, one has
I
tg( ) X1 1
(4.5.10) d = 2i
tg(z ) + 2i [k 2 D℄
z k= 1 k z
D
1
P 1
Proof. As was proved in Le
ture 3.6, series ( z)( k) admits term-wise
k= 1
integration. The residues of 1 are 1 at k and z 1k at z . Hen
e
( z)( k) k z
I (
1 2i z 1k ; for k 2= D
d =
( z )( k) 0; if k 2 D:
D
It follows
I
tg( ) X1 1 X1 1
d = 2i [k 2= D℄ = 2i
tg(z ) [k 2 D℄:
z k = 1 z k k = 1 z k
D
Lemma 10.
tg(z ) is an analyti
fun
tion dened on the whole plane, having
all -integers as its singular points, where it has residues 1.
Proof. Consider a point z 2= Z. Consider a disk D, not
ontaining -integers
with
enter at z . Then formula (4.5.10) transforms to the Cau
hy Integral
Formula. And our assertion is proved by termwise integration of the power
expansion of 1 z just with the same arguments as was applied there. The same
formula (4.5.10) allows us to evaluate the residues.
1
P
Theorem 11.
ot z = z1 + z2 2kz2 2
k=1
Proof. Consider the dieren
e R(z ) =
ot z
tg(z ). This is an analyti
fun
tion
whi
h has -integers
H
as singular points and has residues 0 in all of these. Hen
e
R(z ) = 21i R (z) d for any z 2= Z. We will prove that R(z ) is
onstant. Let
D
z0 and be a pair of dierent points not belonging to Z. Then for any D su
h
that D \ Z = ? one has
(4.5.11) I I
1 1 1 1 R(z )(z z0 )
R(z ) R(z0 ) = R( ) d =
2i z z0 2i ( z )( z0 )
D D
Let us dene Dn for a natural n > 3 as the re
tangle bounded by the lines Re z =
(=2 n), Im z = n. Sin
e jR(z )j 7 by Lemmas 2, 3, 7, and 8 the inte-
grand of (4.5.11) eventually is bounded by 7jzn2z0 j . The
ontour of integration
onsists of four monotone
urve of diameter < 2n. By the Lemma on estimate,
the integral
an be estimated from above by 32n7nj2z z0 j . Hen
e the limit of our
integral as n tends to innity is 0. This implies R(z ) = R(z0 ). Hen
e R(z ) is
on-
stant and the value of the
onstant we nd by putting z = =2. As
ot =2 = 0,
140
n
P Pn
the value of the
onstant is
tg(=2) = nlim 1 2 1
!1 =2 k = nlim
!1 k= n 1 2k .
k= n
n
P P0 n n
This limit is zero be
ause 1 = 1 + P 1 = P 1 +
1 2k 1 2k 1 2k 2k+1
k= n k= n k=1 k=0
n
P 1 = 2n1+1 .
2k 1
k=1
Summation of series by
ot z .
Theorem 12. For any rational fun
tion R(z ), whi
h is not singular in integers
1
P P
and has degree 2, one has R(n) = z res
ot(z )R(z )
1
k=
H
Proof. In this
ase the integral nlim
!1 R(z )
ot z = 0. Hen
e the sum of
Dn =pi
P1
all residues of R(z )
ot z is zero. The residues at -integers gives R(k).
P k= 1
The rest gives z res
ot(z )R(z )
X1
1 1
X1 1 1
(4.5.13) + + + =
k= 1 z k k k=0
z k k+1
X1 1 1
X1 1 1
+ + +
k=1
k z+k k=1
z+1 k k
The rst of the series on the right-hand side represents (z )
, the se
ond is
( z + 1) +
. We get the following
omplement formula for digamma fun
tion:
(4.5.14) (z ) + (1 z ) =
ot z
Sin
e 00 (z + 1) = 0 (z ) = =(z ) (Lemma 11) it follows 0 (1 + z ) = (z ) +
and 0 ( z ) = ( (1 z ) +
). Therefore 0 (1 + z ) + 0 ( z ) =
ot z .
Integration of the latter equality gives. (1 + z ) ( z ) = ln sin z +
:.
Changing z by z we get (1 z ) + (z ) = ln sin z +
. Exponentiating
gives (1 z ) ( z ) = sin1z
. One denes the
onstant by putting z = 21 . On
the left-hand side one gets ( 21 )2 = , on the right-hand side,
. Finally we get
the
omplement formula for Gamma-fun
tion :
(4.5.15) (1 z ) (z ) =
sin z
141
1
Q x2
Now
onsider the produ
t (1 k2 ). Its
anoni
al form is
k=1
1 n
Y x nx o 1 Y1 n x x
o 1
(4.5.16) 1 e 1+ e n
k=1
n k=1
n
x
x
The rst produ
t of (4.5.16) is equal to x e( x) , and the se
ond one is xe ( x) .
Therefore the whole produ
t is x2 (x1) ( x) . Sin
e (1 x) = x ( x) we get
the following result
1 Y1 x2
(4.5.17) =x 1 :
(x) (1 x) k=1 k2
Comparing this to (4.5.15) and substituting x for x we get the Euler formula:
Y1 x2
(4.5.18) sin x = x 1
k=1
2 k2
Problems.
1. Expand tan z into power series
1 1
P
2. Evaluate 1+k2
k=1
1
P 1
3. Evaluate 1+k4
k=1
142
Bibliography
[1℄ E. Hairer, G. Wanner, Analysis by Its Hystory, Undergraduate Texts in
Mathemati
s, Springer, 1997
[2℄ G. H. Hardy, Divergent Series, Oxford University Press, London, 1949
[3℄ R. L. Graham, D. E. Knuth, O. Patashnik, Con
rete Mathemati
s, Addison-
Wesley, 1994
[4℄ L. Euler, Introdu
tio in analysin innitorum, t.1, Lausannae, 1748
143