Sequences and Their Limits: The Limit Idea
Sequences and Their Limits: The Limit Idea
Sequences and Their Limits: The Limit Idea
x1 , x2 , x3 , . . . , xn , . . .
that goes on indefinitely. The numbers in the sequence are usually called terms,
so that x1 is the first term, x2 is the second term, and the entry xn in the general
nth position is the nth term, naturally. The subscript n = 1, 2, 3, . . . that marks
the position of the terms will sometimes be called the index. We shall deal only
with real sequences, namely those whose terms are real numbers. Here are some
examples of sequences.
• the decimal sequence that estimates 1/3: .3, .33, .333, .3333, .33333, . . .
• a binary sequence: 0, 1, 0, 1, 0, 1, . . .
• a geometric sequence: 1, r, r2 , r3 , . . . , rn , . . .
(−1) n
• a sequence that alternates in sign: 21 , −1 1
3 , 4, . . . , n , . . .
• a decreasing sequence: 1, 21 , 13 , 14 , . . . , n1 , . . .
1
• a seemingly random sequence: sin 1, sin 2, sin 3, . . . , sin n, . . .
• a sequence that lists all fractions between 0 and 1, written in their lowest
form, in groups of increasing denominator with increasing numerator in each
group:
1 1 2 1 3 1 2 3 4 1 5 1 2 3 4 5 6 1 3 5 7 1 2 4 5
, , , , , , , , , , , , , , , , , , , , , , , , ,...
2 3 3 4 4 5 5 5 5 6 6 7 7 7 7 7 7 8 8 8 8 9 9 9 9
{xn }∞
n=1 , or more briefly {xn } , or even the unadorned xn
are suitable as well. For some sequences it is possible to give a simple formula for
the nth term as a function of the index n. For example, the nth term of the sequence
1, 1/2, 1/3, 1/4, . . . is xn = 1/n. For other sequences, such as the sequence of
primes or the sequence for the decimal expansion of π, a clean formula for the nth
term is not available. Nevertheless, the entry in the nth position remains uniquely
specified.
At times the sequence {xn } is given, not by a direct formula for the nth term,
but rather recursively. To specify a sequence recursively, you state explicitly what
one or more of the beginning terms are, and then you give a formula for the general
entry in terms of its preceding terms. Here is an example of a famous sequence that
is defined recursively. Let
According to this specification, the first few terms of this sequence {fn } go as
follows:
1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, 233, . . .
This is the celebrated Fibonacci sequence.
2
Notice that the index n need not always start at n = 1. In the Fibonacci se-
quence it has been√ traditional to have the index start at n = 0. For the sequence
given by xn = n − 2 it is suitable to have the index start at n = 2. For our
purposes, it is not so important where the index starts. What matters most in dis-
cussing limits is the behaviour of the sequence in the long run. That is, for large
values of the index.
Exercises
1. In the displayed sequence above, that lists all rational numbers between 0
and 1 in order of increasing denominator, write the next 10 terms after the
displayed term of 5/9.
3
much accuracy we specify, the limit can be approximated to satisfy that accuracy
if we wait long enough.
This quantitative approach brings us to a central idea in calculus. The idea is
that a sequence {xn } has a limit p provided xn can be brought as close to p as we
like by simply going far enough out in the sequence. In the tradition of calculus the
symbol used to specify an arbitrary amount of closeness is the Greek letter , called
epsilon. You should get used to thinking of the letter to represent an arbitrary, yet
very small positive number.
Here is the formal and very important definition of limit of a sequence.
4
Example 1.
n
The sequence given by xn = seems to have the limit p = 1, as n tends to
n+1
∞. Let us pursue this observation in a quantitative manner as prescribed by the
definition of limit. Thus take an arbitrary tolerance > 0. Now we must find a
cut-off number K such that
n
n + 1 − 1 < whenever n > K.
Using a common denominator, this boils down to finding a number K such that
−1
n + 1 < whenever n > K.
With a little more algebra it becomes apparent that we need to find a number K
such that
1
< n + 1 whenever n > K.
In other words we need a number K such that
1
− 1 < n whenever n > K.
1
Such a cut-off K is now apparent, namely take K = − 1. We can be sure that
1 n
if n > − 1, then − 1 < .
n + 1
For instance, suppose = .002 is given. We have decided above that a suitable
cut-off point is
1
K= − 1 = 499.
.002
In other words, starting with the 500th term of the sequence, you know that from
then on the sequence will be less than the distance .002 away from the limit 1. Try
checking with your calculator that the distance between the limit 1 and the terms
500 501 502 503
, , . ,...
501 502 503 504
is indeed less than .002.
Example 2.
5
√
Take the sequence xn = 1/ n whose limit appears to be p = 0. Say we want
1
|xn − 0| < .
13
What is a good cut-off K which will ensure that that |xn − 0| < 1/13 will indeed
√
happen once n > K? Well, we want 1/ n < 1/13. By squaring we see that this
will happen when 1/n < 1/169, which happens when n > 169. A suitable cut-off
we are looking for is K = 169.
Next say we wanted
1
|xn − 0| < .
100
By the same argument as above we can see that once n > 1002 = 10000, then we
will have |xn − 0| < 1/100. Again a suitable cut-off is available.
More generally if we had any > 0 and we wanted |xn − 0| < , how far
should we take n to be sure that this accuracy of estimation kicks in? Since we
√
want 1/ n < we had better have 1/n < 2 . In other words we had better have
n > 1/2 . A suitable cut-off is K = 1/2 .
At this point somebody might ask:
Regarding the inequality |xn − 0| < 1/13 up above, I can see that
K = 169 is a good cut-off, while K = 168 is not quite good enough.
So 169 seems to be the best possible cut-off that lets us achieve an
accuracy of 1/13 in this example. Do I always have to find the best
possible cut-off as we did in this example?
As far as the definition of limits is concerned, the answer is no. For instance, cut-
offs such as K = 170 or K = 500 are just as suitable. Once a suitable K is
found, any larger K is just as suitable in fulfilling the limit definition. Thus there
is not a “one and only” answer for a suitable cut-off point K. Depending on the
problem, it may be too difficult to determine the best possible cut-off value K. On
the other hand a suitable cut-off may well be obtainable. The limit concept can
tolerate such a compromise. In the next example we show how a suitable cut-off
K can be found, without having to worry about the best possible K.
Example 3.
6
To prove that 0 is the limit, take any > 0. We need to find a cut-off K such that
√ √
n + 1 − n − 0 < when n > K.
It is not so clear how we are to go about isolating n in this inequality. Rather than
confront that difficulty we do something clever. We notice that
1 1
√ √ < √ , since the denominator on the left is bigger.
n+1+ n n
Exercises
5. Take the constant sequence 2, 2, 2, 2, . . . . The limit had better be 2. How far
do you have to go with the terms xn = 2 of the sequence in order to be sure
that |xn − 2| < 10−10 ? The answer should be obvious.
1000 n 1 1
6. Show that if n > , then − < .
9 3n + 1 3 1000
1 n 1
You are given an > 0. If n > , show that − < .
9 3n + 1 3
7
7. Find a cut-off K that ensures 1/2n < when n > K.
8. We think that e−n → 0. Given > 0 find a cut-off K that ensures e−n <
when n > K.
1 2
11. We can sense that 3 + → 9 as n → ∞. If n > 70, show that
n
(3 + 1 )2 − 9 < 1 . Hint: you know that 1 ≤ 1 always.
n 10 n2 n
7 1 2
More generally take any > 0. If n > , show that (3 + ) − 9 < .
n
2
1
This proves that 3 + → 9.
n
n2 − 1 1
12. Apply the limit definition to prove that lim = .
n→∞ 2n2 + 3 2
Here the word ”eventually” captures the idea that there is a cut-off point K that is
suitable for the given , but we would prefer not to name a specific K at this time.
Here come some examples illustrating this less formal language.
Example 4.
8
Of course, this is the same as getting
n
2
< eventually.
3
After dividing by the negative number ln(2/3), we come down to showing that
ln
n> eventually.
ln(2/3)
Now there is no doubt that n will get past the fixed number ln / ln(2/3) eventually.
After all, n goes beyond any fixed number eventually. Thus we have given the proof
that (−2/3)n → 0.
Observe in Example 4 that if we wanted a suitable cut-off, we could take it
to be K = ln / ln(2/3). For instance, when = 1/106 is the given tolerance, a
suitable cut-off number is
So we learn that once our sequence hits the 35th term, from then on it will estimate
0 with an accuracy that is better than 1/106 .
Example 4 can readily be modified to prove that
There was nothing special about the base r = −2/3, other than the fact it lies
between −1 and 1.
Example 5.
9
What this really means is that
3 33 333 3333 1
the limit of the sequence , , , , . . . is .
10 100 1000 10000 3
333 · · · 3
If xn = where each 3 and each 0 is repeated n times, let us show that
1000 · · · 0
indeed xn → 1/3. Given a small tolerance > 0 we must prove that
xn − 1 < eventually.
3
and next
x2 − 1 33 1 99 100 1
= − = − = ,
3 100 3 300 300 300
and following this pattern
xn − 1 1
= .
3 3 · 10n
So we need to know that
1
< eventually.
3 · 10n
In other words we need to know that
1
< 10n eventually.
3
It should now be quite clear that that 10n goes past any number eventually, so we
are done. If we wanted to actually find a suitable cut-off K, we can go one step
further and observe that the last inequality amounts to having
1
log10 < n eventually.
3
We can see that once n has gone beyond the cut-off number K = log10 (1/3),
then |xn − 1/3| < will kick in.
Example 6.
10
This example is a bit more subtle. By testing the values of cos(1/n) on a calculator
(using radian mode!) it becomes evident that this sequence tends to 1. Let us verify
that
1
cos → 1 as n → ∞.
n
Take any tolerance > 0. Now we need to show that
cos 1 − 1 < eventually.
n
After we remember that 1 = cos(0) the problem comes down to showing that
cos 1 − cos(0) < eventually.
n
We need to do a bit of trigonometry on the side. Namely, observe that for any two
angles α and β (in radian mode) we have the inequality
11
This significant little fact can be seen from the diagram that follows.
..............................................
....................
...............
.............
............
..........
.......... R
.........
........
........
........
.......
.......
.......
.......
.......
......
......
......
.....
....
α−β
....
....
....
....
....
....
...
...
...
...
...
...
...
...
S Q ...
...
...
...
...
...
...
...
...
...
...
..
..
..
..
..
..
..
β
..
..
..
..
..
..
..
..
..
O A B P = (1, 0)
β = arclength P Q OB = cos β
α = arclength P R OA = cos α
α − β = arclength QR ≥ SQ = AB = cos β − cos α
12
Exercises
n
13. Show that < .001 eventually.
n2
+1
n
For any > 0 show that 2 < eventually.
n +1
14. Given > 0, show that sin(1/n) < eventually.
Hint: | sin x| ≤ |x| for all numbers x.
√
15. Prove that sin(n2 )/ n < 10−9 eventually. Does it still work if we replace
10−9 by any > 0?
16. Prove that 1/10n → 0 by showing that for any > 0 we have 1/10n <
eventually.
1 1
17. Show that ln e +
− 1 < eventually. Repeat the problem with
n 100
1
> 0 replacing . Hint: 1 = ln e.
100
18. Given > 0, show that 2n /n! < eventually. This is a bit harder. First look
at and explain the hint:
2n 2 2 2 2 2 2 4
= · · · · ··· · ≤ .
n! 1 2 3 4 n−1 n n
You have just proven that 2n /n! → 0. Next prove that 3n /n! → 0 by the
method of imitation of what you just did with 2n /n!. Imitate some more and
prove 10n /n! → 0. Given any constant a, what do you think an /n! tends
to?
19. Show that for any > 0 we have n!/nn < eventually.
Hint: first show that n!/nn ≤ 1/n.
20. Prove that n/en → 0 by showing that for any > 0 you will eventually get
n/en < .
Hint: You can see that n/en = (n/2n )(2/e)n ≤ (2/e)n since n/2n ≤ 1.
21. This is a little bit of theory. If a sequence tends to a positive limit, show that
the sequence eventually must itself become positive. More precisely suppose
that
xn → p and that p > 0.
Prove that xn > 0 eventually.
13
Sequences that do not converge
Of course not all sequences have a limit. Let us examine a couple of situations of
this sort.
Example 7.
14
Bounded sequences
The sequence in Example 8 has the property that it is not bounded. A sequence
{xn } is called bounded provided that there are two constants a, b which satisfy
The constants a, b are called lower and upper bounds of the sequence, respectively.
A simple example of a bounded sequence is
1, 0, 1, 0, 1, 0, 1, 0, . . .
Here is the explanation of this fact in terms of the limit definition. Let the sequence
{xn } converge to the limit p. Taking = 1 we know that |xn − p| < 1 eventually.
Unfolding this inequality we get that p − 1 < xn < p + 1 eventually. So the terms
are eventully stuck between two bounds. In other words, there are only a finite
number of terms not stuck between two bounds. By lowering the lower bound and
raising the upper bound to take in the few terms that are missing, we can make sure
that all the terms are stuck between two bounds. Thus every sequence with a limit
is bounded.
Knowing that every convergent sequence is bounded, we thereby know that
an unbounded sequence has no limit. This little fact can be used to rule out the
presence of limits in some cases. Thus, for example, sequences such as xn = log n
√
or xn = 2n or xn = n have no limit because they are not bounded. However be
careful! Whereas unbounded sequences never converge, a bounded sequence may
or may not converge.
Exercises
22. Show that the sequence xn = ln n is not bounded by proving that every
number b fails to be a bound.
Note: in order that b not be a bound for the sequence, you have demonstrate
that there is some n so that ln n > b.
Does the sequence ln n converge?
15
n3 + 1
23. Show that is eventually greater than any given number b. Why does
n2 − 5
this sequence have no limit?
n3 + 1 n3 + 1 1
Hint: first explain why 2 > 2
= n + 2 , then use this fact.
n −5 n n
24. Is the sequence cos(nπ/4) is bounded? Does this sequence have a limit?
Explain your answers.
16
Theorem 9. If xn → p and yn → q as n → ∞, then
• xn + yn → p + q (addition formula)
• xn − yn → p − q (subtraction formula)
• xn yn → pq (multiplication formula)
xn p
• → , provided that yn 6= 0 and q 6= 0. (division formula)
yn q
Proof. Although these properties may seem obvious, we can only be certain that
they are correct by putting them to the test of the limit definition. What is difficult
to appreciate about this theorem is, not so much the proof, but rather the fact some-
thing needs to be proved. Let us prove just the addition formula, that the limit of
a sum is the sum of the limits, and take the other three properties on faith. We are
given that
xn → p and yn → q as n → ∞.
We must prove that
xn + yn → p + q as n → ∞.
Take > 0. According to the limit definition we must show that
17
Example 10.
2n3 − 5n + 7
Let us find the limit of . We are tempted to use the division formula
8n3 + 9n2 − 4
for limits. Unfortunately both the numerator 2n3 − 5n + 7 and the denominator
8n3 + 9n2 − 4 blow up for large values of n, and thereby do not have a limits to
which the division formula can be applied. We need a trick! Divide both numerator
and denominator by the highest power of n, namely n3 , and you get
Now use Theorem 9 liberally and repeatedly. We know the constant sequence 7
tends to the limit 7, while the sequence 1/n3 tends to 0. By the multiplication
formula 7/n3 → 7 · 0 = 0. Likewise 5/n2 → 0. Clearly the constant sequence 2
has limit of 2. Using the addition and subtraction formulas, 2 − 5/n2 + 7/n3 → 2.
Likewise, by repeated use of the addition, subtraction and multiplication formulas,
we obtain 8 + 9/n − 4/n3 → 8. Next by the division formula we conclude that
5
2− n2
+ n73 2 1
→ = .
8 + n9 − n43 8 4
Note. Once a trick such as the one above is used, it is no longer a trick. You
now have a method. You should be able to adapt the method of this example to
numerous sequences of this sort. Normally, you do not have to provide an alert
every time a limit property from Theorem 9 is being used. It is OK to say
5
2− n2
+ n73 1
→ by inspection,
8 + n9 − n43 4
but keep in mind that, in the background, you are using Theorem 9 repeatedly.
Now comes a theorem about sequence limits that relates to function continuity.
Theorem 11. Let f be any function that is continuous at some point p. That is
f (x) → f (p) as x → p or lim f (x) = f (p). We know a multitude of such
x→p
functions. If xn is a sequence in the domain of f and xn → p, then f (xn ) → f (p).
Proof. The proof is just a simple matter of keeping track of the limit definitions,
including the -δ definition for continuity of a function.
18
Take any < 0. We need a cut-off number K such that
When n > K, our terms xn are exactly where they need to be in order to guarantee
that |f (xn ) − f (p)| < .
19
sin x
For an illustration of Theorem 12, notice that → 1 as x → 0. We also
√ x
know that 1/ n → 0 as n → ∞. By Theorem 12
√
√ √ sin (1/ n)
n sin 1/ n = √ → 1 as n → ∞.
1/ n
Here is an example of the easy and somewhat dull ways in which Theorems 9,
11 and 12 get used together.
Example 13.
√
Let us show that 71/2+1/n tan((nπ + 1)/4n) → 7.
20
Exercises
√
25. Find the limit of n2 + n − n. It is not required to use the definition of
limits, just do some algebraic manipulations and make an inspection based
on Theorems 9 and 11.
p
26. Find lim n2 − 2n − n, by putting this sequence into a form that accepts
n→∞
the rules of Theorems 9 and 11.
One can prove that this sequence converges, say to p. Use Theorem 11 to
show that (p, cos p) is the point where the graph of y = cos x intersects the
line y = x.
q
28. Find the limit of n 1 + n1 − 1 by putting this sequence into a form that
Theorems 9 and 11 can digest.
q
3 1
29. Find the limit of n 1 + n − 1 by putting this sequence into a form that
Theorems 9 and 11 can digest. This one is a bit trickier.
1
30. For each real number a consider the sequence xn = . Depending on
1 + an
a you may get a different limit or none at all. Use your powers of inspection
to decide which a’s give which limits.
32. Find limn→∞ n/ ln(1 + 2en ), showing how you got it.
21
variable is the integer n that runs through the discrete values 1, 2, 3, . . . . This
technical distinction needs to be kept in mind in order to avoid mixups.
For example, take the function y = sin(πt). As t → ∞ the function sin(πt)
has no limit since it oscillates between ±1. However, the sequence xn = sin(πn)
defined for n = 1, 2, 3, . . . is the same as the zero sequence 0, 0, 0, . . . , and the
limit of this is surely 0.
Thus limits of sequences do not help us determine horizontal asymptotes. Hap-
pily, horizontal asymptotes do help us determine limits of sequences.
Informally speaking, Proposition 14 is true simply because the n’s are just
particular t’s.
For instance we can use L’Hôpital’s rule to get that t2 /et → 0 as t → ∞. Thus
the sequence n2 /en → 0 as n → ∞.
Exercises
try computing the first 50 or so terms of this sequence, using a calculator. Us-
ing radian mode punch in cos 1. Then take the cos of the output, then take the
cos of that, then the cos of that etc. You quickly see a limit emerging as the
outputs begin to stabilize. That limit p is a new number with no name. In fact
p ≈ 0.7390851332151607. Suppose xn → p. Then cos xn → cos p. On the other
hand cos xn = xn+1 and xn+1 → p. We get that cos p = p. Our recursive little
sequence has just solved the difficult equation cos x = x.
22
For such exotic new numbers it is the sequence itself that provides the mecha-
nism for estimating and thereby understanding the limit. If we have no clue what
the limit of a sequence could be, the best we could ask for is to decide whether
or not a given sequence simply has a limit. From this point of view, a sequence
becomes a machine that uncovers previously unknown numbers.
When we do not know the limit in advance, the most basic method for deciding
that a sequence has a limit is the so called monotone convergence principle or the
monotonic sequence theorem. We will now discuss that principle.
We say that a sequence {xn } is increasing provided
x1 ≤ x2 ≤ x3 ≤ · · · ≤ xn ≤ xn+1 ≤ . . .
If the inequalities are strict (i.e < instead of ≤) we say that the sequence is strictly
increasing. Naturally, if the inequalities go the other way the sequence is called
decreasing. A sequence that is either increasing or decreasing is known as a mono-
tonic sequence.
Recall that a sequence {xn } is bounded provided that all terms satisfy
called lower and upper bounds respectively. Bounded sequences need not con-
verge, for instance take the bounded sequence 0, 1, 0, 1, 0, . . . which does not con-
verge. Monotonic sequences need not converge, for instance take the increasing
sequence 1, 2, 3, 4, . . . . However a sequence with both these two properties to-
gether (i.e. bounded and monotonic) always has a limit. Let us now state and
prove this important result.
Proof. Let’s say the sequence is increasing. Since the sequence is bounded there
has to be an upper bound. Of all possible upper bounds for the xn terms, let p be
the smallest one. Thus p is a number with the fundamental property that
23
t t
x1 x2 x3 ··· xn xn+1 ··· p b
It turns out that p is our limit. To see that, we go to the limit definition. Take any
> 0. We need to show that
p − < xK ≤ xn ≤ p < p + .
|xn − p| < once n goes past the cut-off K, which is to say, eventually.
A similar proof, based on taking greatest lower bounds, works for decreasing
sequences.
Here come some examples of bounded, monotone sequences and their limits.
Example 16.
24
A preliminary inspection might suggest that this sequence is increasing. Furthe-
more it seems to not be growing too fast. Maybe it is bounded. After some sam-
pling we suspect that the sequence stays below a bound such as 4. Let use the
method of mathematical induction to show that
We can build up to what we want from our previous inductive assumption. Since
we deduce that
3 + 2xn < 3 + 2xn+1 < 3 + 2 · 4 = 13.
Hence √ p √
3 + 2xn < 3 + 2xn+1 < 13.
Taking into account how xn was built recursively we have just discovered that
√
xn+1 < xn+2 < 13 < 4,
25
Example 17.
1 1 1 1 1 1
1, 1 + 1, 1 + 1 + , 1 + 1 + + , 1 + 1 + + + ,
2 2 6 2 6 24
1 1 1 1 1 1 1 1 1
1+1+ + + + , 1+1+ + + + + ,...
2 6 24 120 2 6 24 120 720
which simplify to
5 8 65 163 1957
1, 2, , , , , ,...
2 3 24 60 720
You can check with your calculator that these numbers seem to approximate the
decimal expansion for the special number e. In fact this sequence does tend to e.
This sequence becomes a decent mechanism for using fractions to estimate e to
any desired level of accuracy.
We will now satisfy ourselves that this sequence converges by using the mono-
tonic sequence theorem. After starting with x0 = 1, each term of the sequence is
obtained by adding a positive amount to the previous term. So there is no doubt
that the sequence is increasing. In order to be sure the sequence converges, all we
need to do is check that the sequence is bounded. For sure 0 ≤ xn . To get an upper
bound notice that for every positive integer k we have
k! = 1 · 2 · 3 · · · (k − 1) · k ≥ 2| · 2 ·{z2 · · · 2} = 2k−1 .
k−1 2s here
1 1
Thus ≤ k−1 , and therefore
k! 2
1 1 1 1 1 1 1
xn = 1 + 1 + + + ··· + ≤ 1 + 1 + + 2 + 3 + · · · + n−1 .
2! 3! n! 2 2 2 2
After recalling the sum formula for a geometric series we see that
1 − 21n
1 1 1 1 1
xn ≤ 1 + 1 + + 2 + 3 + · · · + n−1 = 1 + 1 ≤1+ 1 = 3.
2 2 2 2 1− 2 1− 2
26
Some mathematicians make the declaration that e, by definition, is the limit of
the sequence in Example 17. If e is thus defined, we should notice from the proof
of the monotonic sequence theorem that e is the smallest of the upper bounds for
xn . We saw in the proof that 3 is an upper bound for xn , and we can see that 2
is not an upper bound for xn , since x3 is already past 2. Therefore the smallest
upper bound has to be bigger than 2 and no bigger than 3. In this way we come to
understand why the crude estimate 2 < e ≤ 3 is true.
Example 18.
27
Example 19.
You may have wondered how your calculator implements the square root function:
√
. What it does, is generate a simple numerical sequence with the desired square
root as its limit. The example to follow illustrates the mathematics that could well
be going on inside your calculator. This process, typically known as an algorithm,
deserves our attention at least for the sake of its practicality.
We propose√ to build a bounded, monotonic sequence of ordinary
√ fractions that
converges to 10. This will give us a method of computing 10 to any desired
accuracy. Here is our sequence defined recursively:
1 10
x1 = 10, xn+1 = xn + for n = 1, 2, 3, . . . .
2 xn
For instance, the first few terms of the sequence are:
11 161 45281 4057691201
10, , , , ,....
2 44 14168 1283082416
Using the monotone convergence principle we can show that this sequence con-
verges. First notice that 0 < xn , just by inspecting how each successive term of the
sequence is built. Using some careful algebra let us now show that this sequence
decreases. To that end we must prove xn+1 ≤ xn for n
= 1, 2, 3, . . . . Given how
1 10
we built xn+1 , we are obliged to show that xn + ≤ xn . Because 0 < xn ,
2 xn
you should check that this simplifies down to proving
For n =1 we obvioulyhave 10 ≤ x21 = 100, while for n > 1 we know that
1 10
xn = xn−1 + . Now watch closely:
2 xn−1
10 2
2 1
xn = xn−1 +
4 xn−1
1 2 100
= xn−1 + 20 + 2
4 xn−1
1 100 20 20
= 10 + x2n−1 − 20 + 2 , because = 10 −
4 xn−1 4 4
2
1 10
= 10 + xn−1 −
4 xn−1
≥ 10, since the term dropped is a square and thus ≥ 0.
28
So 10 ≤ x2n , and our explanation above shows that xn decreases. Since 0 < xn
our sequence has a lower bound, and because it decreases it certainly has an upper
bound, namely the first term 10. The monotone convergence principle kicks in!
The sequence xn has a limit, call it p.
What could the limit p possibly be? Because xn → p we get that xn+1 → p
also. After all, xn+1is just theoriginal sequence with the first term dropped.
1 10 1 10
However xn+1 = xn + . Consequently xn+1 → p+ , using
2 xn 2 p
1 10
the familiar limit properties of Theorem 9. We conclude that p = p+ .
√ 2 p
Solving√this equation using simple algebra we get p = ± 10. It cannot be that
p = − 10 √ since the sequence xn is√ always positive, and therefore gets nowhere
near to − 10. It must be that p =√ 10. We have therefore built a sequence of
ordinary fractions that converges to 10.
√
The method illustrated here for finding 10 can readily be adapted to finding
√
a for any a > 1. You simply replace the 10 in the definition of xn by the a that
you have in mind. What is truly remarkable about this sequence-based algorithm
for finding square roots is the fantastic speed at which the recursive sequence con-
verges. It turns out that after the recursion gets going a little bit, the number of
√
decimal places of accuracy in √ estimating a will roughly √ double as we pass from
xn to x√n+1 . To illustrate with 10, it turns√out that x4 ≈ 10 to 1 decimal place,
x5 ≈ 10 to 3 decimal places, and x6 ≈ 10 to at least 7 decimal places. This
kind of ultra fast convergence is sometimes known as quadratic convergence. What
makes your calculator work so well is not just the high speed electronics, but also
the intelligent mathematics that the programs within it exploit.
Exercises
x1 = 0, xn+1 = 1 − xn for n ≥ 1.
29
36. Let {xn } be the sequence defined recursively by
1
x1 = 2 and xn+1 = (xn + 6) for n = 1, 2, 3, . . . .
2
Show that this is an increasing and bounded sequence. Find its limit.
Hint: 10 should work as an upper bound.
n
37. We can see that converges, to 0 in fact. Use the monotone sequence
n2
+1
theorem to show this sequence converges by proving that xn > xn+1 and
that xn is bounded below by some number.
39. We can see that the sequence (2n − 3)/(3n + 4) converges to 2/3. Thus
it must be bounded. Decide if the sequence monotonic by sampling a few
values, then making a decision, and then proving your decision.
40. Apply the monotone convergence principle to show that the sequence given
recursively by
1
x1 = 2 and xn+1 = for n = 1, 2, 3, . . . .
3 − xn
converges. Then find its limit.
Since the sequence is increasing and bounded, it has limit p. Show that p is
a solution of the equation x4 − 4x2 − x + 4 = 0. By plugging in the value
1, we see that 1 is also a solution of this equation. Explain why p 6= 1. By
factoring x − 1 from x4 − 4x2 − x + 4, find a cubic polynomial that has p as
its root.
30
42. Explain why the sequence xn = arctan n converges. Find its limit and give
an intuitive reason for your answer.
Show that this sequence increases. If this sequence is bounded, what must
the limit be? Explain why such a limit is impossible. Is this sequence
bounded?
√
44. Imitate the algorithm of Example 17 to estimate 50 as a fraction. Keep go-
ing√with your algorithm until your estimate agrees with the calculator value
of 50 to 3 decimals.
31