Lecture 33
Lecture 33
Lecture 33
I Introduction to series
I Error in linear approximation
I Taylor polynomials
Definition 1
Given a sequence of real numbers {an }∞
n=1 , we define the series as the formal expression
a1 + a2 + a3 + ... + an + ...
∞
X
and denote this by an .
n=1
The nth term of the sequence, an , is also referred to as the nth term of the series.
The nth partial sum is Sn = a1 + a2 + ... + an and the sequence of partial sums is the sequence {Sn }∞
n=1 .
We say that the series converges if the sequence of partial sums converges and we say that the series diverges if
the sequence of partial sums diverges.
If the series converges, then we define its sum to be the limit of the sequence of partial sums and use the same
notation for the sum as for the series itself, i.e.,
∞
X
an = a1 + a2 + a3 + ...an + ... = lim Sn .
n→∞
n=1
Examples of divergent series
Consider the sequence {(−1)n }∞ n
n=1 . Does the series −1 + 1 − 1 + 1 − ... + (−1) + ... converge or
diverge?
If we write this as
(−1 + 1) + (−1 + 1) + (−1 + 1) + ...,
we would expect the sum to be 0 + 0 + 0 + ... = 0.
We could also write the series as
−1 + (1 − 1) + (1 − 1) + (1 − 1) + ...,
The starting index of the sequence (and hence, the series) does not have to be 1. E.g., we can define
series
X∞ X∞
an = a0 + a1 + ... + an + ..., or an = a10 + a11 + ... + an + ...
n=0 n=10
∞
X
The nth partial sum of an is the sum of the first n terms, i.e., Sn = aN + aN+1 + ... + aN+n−1 for
n=N
n ≥ 1.
Proposition
Let N and M be any two integers such that an is defined for all integers n ≥ min{N, M}. Then, the series
X∞ ∞
X
an converges if and only if an converges.
n=N n=M
P
If it is clear from the context or not important to specify the starting index, we may use the notation an as
shorthand for the series.
∞
X ∞
X
Also, we can freely change the name of the index variable, e.g., an = ak .
n=1 k=1
Geometric series
Definition 2
The geometric series with common ratio r ∈ R is the series
∞
X
r n−1 = 1 + r + r 2 + ... + r n−1 + ....
n=1
Note: It is common to write geometric series starting from index 0, in which case the expression is
∞
X
r n.
n=0
Proposition 1
The geometric series converges if |r | < 1 and diverges if |r | ≥ 1.
∞
X 1
For |r | < 1, the sum is r n−1 = .
n=1
1−r
This theorem essentially says that the sequence {Sn } converges if and only if it is Cauchy.
The alternating harmonic series - part 1
∞
X (−1)n+1
Claim: The alternating harmonic series converges.
n=1
n
Proof: We will show that the sequence of partial sums is Cauchy (the Cauchy condition).
Suppose m and n are integers with m > n ≥ 1. Then,
m
(−1)m+1 (−1)n+1
X 1 1 1 1 1 1
|Sm − Sn | = ak = 1− + − + ··· + − 1 − + − + ··· +
k=n
2 3 4 m 2 3 4 n
(−1)n+2 (−1)n+3 (−1)n+4 (−1)m (−1)m+1
= + + + ··· + +
n+1 n+2 n+3 m−1 m
(−1)m−n−2 (−1)m−n−1
n+2 1 1 1
= (−1) − + − ··· + +
n+1 n+2 n+3 m−1 m
1 1 1 (−1)m−n−2 (−1)m−n−1
= − + − ··· + + .
n+1 n+2 n+3 m−1 m
The alternating harmonic series - part 2
m
X 1 1 1 1 1 1
ak = − + − +··· + − .
k=n
n+1 n+2 n+3 n+4 m−1 m
| {z } | {z } | {z }
>0 >0 >0
We can remove the absolute value sign on the right-hand side since it is a sum of positive terms (when grouped
as indicated above).
Grouping in a different way, we have
m
X 1 1 1 1 1 1 1 1
ak = − − − ··· − − − < −
k=n
n+1 n+2 n+3 m−2 m−1 m n+1 m
1 1
since the neglected terms − n+2
− n+3
− . . . are all negative. Furthermore, we see that
Pm 1
k=n ak < n+1 .
Pm
Given any ε > 0, choose N > ε1 . Then if m > n ≥ N, and m − n is even, we have k=n ak < 1
n+1
< ε, as
required. It just remains to check the case m − n odd (exercise).
Error in linear approximation
Recall linear approximation: Lfa (x). We previously showed that the error approaches 0 “faster than linearly” as x
approaches a. Now, we can look for error bounds at specific points x 6= a.
Start of proof:
For simplicity, suppose I = R. We must show that:
M
1. f (x) ≤ Lfa (x) + 2
(x − a)2 for all x ≥ a
M
2. f (x) ≥ Lfa (x) − 2
(x − a)2 for all x ≥ a
M
3. f (x) ≤ Lfa (x) + 2
(x − a)2 for all x < a
M
4. f (x) ≥ Lfa (x) + 2
(x − a)2 for all x < a
We prove Case 1 on the next slide. The others are essentially the same.
Proof
From the theorem on “comparison of functions by their derivatives” in Lecture 29, if there exists a function g
such that f (a) = g (a) and f 0 (x) ≤ g 0 (x) for all x > a, then
Applying the theorem to the derivatives, if f 0 (a) = g 0 (a) and f 00 (x) ≤ g 00 (x) for all x > a, then
By hypothesis, f 00 (x) ≤ M for all x > a. We seek a function g such that g 00 (x) = M.
Thus, g 0 is an antiderivative of the constant function M, so we have the form g 0 (x) = Mx + C .
Requiring g 0 (a) = f 0 (a) specifies the constant C = f 0 (a) − Ma so we have g 0 (x) = f 0 (a) + M(x − a).
The function g is an antiderivative of g 0 and we want g 0 (a) = f 0 (a) so we obtain
M
g (x) = f (a) + f 0 (a)(x − a) + (x − a)2 .
2
The comparison theorem gives
M
f (x) ≤ g (x) = Lfa (x) + (x − a)2 for all x ≥ a.
2
Better than linear approximations
Qaf (x) = c2 x 2 + c1 x + c0 .
How should we choose c1 , c2 , and c3 ? What properties should the quadratic approximation satisfy?
We should find that Qaf (x) = f (a) + f 0 (a)(x − a) + 21 f 00 (c)(x − a)2 has the properties:
Qaf (a) = f (a), (Qaf )0 (a) = f 0 (a), and (Qaf )00 (a) = f 00 (a).
Taylor polynomials
This idea can be extended to polynomials of higher degree. Such functions are called Taylor
polynomials.
Definition
If f is n-times differentiable at x = a, then the Taylor polynomial of degree (or order) n for f centered at a is
the polynomial
n
X f (k) (a)
Tn,a (x) = (x − a)k
k=0
k!
f 00 (a) f (n) (a)
= f (a) + f 0 (a)(x − a) + (x − a)2 + · · · + (x − a)n .
2 n!
The notation f (k) refers to the kth derivative of f for k ≥ 1 and f (0) = f .
The formula can be repeatedly differentiated by the Power Rule to verify that:
(k)
Tn,a (a) = f (k) (a) for all 0 ≤ k ≤ n
Exercise
Exercise
Find the Taylor polynomial T2,0 (x) for the function f (x) = 3x 3 − x + 5.
Taylor remainders and errors
Definition
If f is n-times differentiable at x = a and Tn,a is its Taylor polynomial of degree n for f centered at a, then the
Taylor remainder function of degree n for f centered at a is defined by
Remarks:
1. The error depends on x. It is possible to have 0 error at some points and large errors at other points.
2. En,a (a) = 0 by construction.
3. Having a small error at one point x 6= a does not guarantee a good approximation everywhere between a
and x.
4. How the error depends on x and n determines how good the approximation is.