Section 11.
3: The Integral Test
Most of the series we have looked at have either diverged or have converged and we have been able to nd what they converge to. In general however, the problem is much more dicult than what we have seen. Convergence can be broken up into two problems 1) Does a series converge? and 2) What does it converge to? To tackle the rst problem, we shall develop a whole new toolbox based on results we have already used in Calculus. The second problem is much more dicult and requires more advanced mathematics than we currently have. Therefore, our focus for most of this chapter will be on the rst problem (though we shall occasionally use the second). 1. Using Integrals to Determine Convergence - an Example We start with an example. Example 1.1. Show that the sum
n=1
1 n2
converges. Consider the function f (x) = 1/x2 which coincides with this sum for integer values. We can approximate the area under this curve using a Riemann sum. If we take the right hand Riemann sum with subdivisions of length 1, then we get the following innite sum
n=1
1 n2
which is the same as the series we are trying to nd. If we ignore the rst term of the sum and just consider the innite sum
n=2
then the value of this sum will be strictly bounded by the area under the graph from 1 to (since the right hand sum produces an underestimate). But this area can be estimated using an improper integral. Summarizing, we have
n=1
1 1 =1+ 2 n n2 n=1
1
1+
1
1 dx. x2
In particular, the sum will only converge if the integral converges. Calculating, we have 1 1 = 1. dx = x2 x 1 1 Therefore, since the interval converges and is larger than the sum, the sum must converge. Hence the sum
n=1
1 n2
converges. WARNING: Just because it converges does not mean we have found the actual value of the sum. In fact, all we have shown is that the sum
n=1
1 1 =1+ <1+ 2 n n2 n=2
1 dx = 2. x2
2. The Integral Test The method we used in our example approximating the sum with an integral can be applied to lots of other cases to show that a series converges (observe that we did not nd what the sum was (it is in fact 2 /6)). There are just a couple of things which are necessary for it to work: (i ) The sequence in the series must be positive for all values past a certain point. (ii ) The sequence in the series must be decreasing after a certain point (to guarantee that we will be getting an underestimate). (iii ) The function corresponding to the sequence is continuous (so we can take the Riemann sum). If these things are satised, then we can apply the following: Result 2.1. (The Integral Test) Let an be a series with positive terms and let f (x) be the function that results when n is replaced by x. If f is decreasing and continuous on [a, ) then an and
f (x)
a
either both converge or diverge.
Observe that we only start at a - this was illustrated in the example because if we tried to integrate from 0 to , the integral will have diverged. Of course, the key point is that the rst few terms will not aect divergence or convergence - it is the ultimate behavior which counts and this is measured by the integral. We illustrate the power of the integral test with a few examples. Example 2.2. Show that the harmonic series diverges. Observe that the harmonic series 1 n agrees with the function f (x) = 1/x. However, 1 dx = ln(x) x 1 1 which diverges. Since the integral diverges, the corresponding series must diverge.
Example 2.3. Show that the sum n3 converges. First observe that since n 1. Therefore 1 n3 + n 1 +n 1 n3
1 1 . n3 n3 + n Using the integral test, we have 1 1 1 = , dx = 2 3 x 2x 1 2 1 so 1 n3 converges, and hence so does 1 . 3+n n Example 2.4. We dene a p series to be the series 1 np where p is some integer. Find all values of p for which this series converges. Clearly if p 0, then this series diverges (because we shall be adding innitely many numbers greater or equal to 1 together. Also, the functions will not be decreasing for p < 0, so we could not apply the integral
test anyway! Also, the case p = 1 was considered above (it is the harmonic series). Therefore, we may assume that p > 0 and p = 1. The corresponding integral will be
1
1 dx = xp
xp dx = lim
x1p k 1 p
k 1
= lim
k 1p 1 k 1 p 1p
The only part of this limit which could diverge is the term k 1p , and this will diverge if and only if 1 p > 0 or p < 1. Therefore, a p-series 1 np converges if and only if p > 1. Example 2.5. Determine which values of p the series enp converges. If p 0, this clearly diverges, so we assume p < 0. This is a decreasing and continuous function, so we apply the integral test:
exp dx = lim
1 k
exp p
= lim
1 k
ekp ep ep + = p p p
This integral converges for all p > 0, so the series converges for all p > 0. 3. Estimating the Sum of a Series Though out main concern is determining whether a series converges, we shall briey discuss how we can approximate the value of a series which does converge using the integral test. If a series converges, then the more terms we add up, the closer the value will be to the actual value of the sum. Suppose that the actual value of a series is S and the nth partial sum is denoted by Sn . Then we dene the remainder Rn = S Sn , the dierence between the actual value of the series and the partial sum. Obviously, the smaller the remainder is, the closer the partial sum is to the actual value of the series. We can use integrals to determine the remainder: Result 3.1. Suppose f (n) = an where f is continuous, positive, decreasing for x n and an is convergent. If Rn = S Sn , then
f (x)dx
n+1
Rn
n
f (x)dx.
This result basically tells us that the remainder can be approximated by these two integrals. We illustrate. Example 3.2. Find the value of n which guarantees that the remainder will be within 0.1 of the actual value for the sum 1 . n2 Observe that 1 1 1 = dx = 2 x x n n n If we choose n = 10, then we have 1 1 R10 11 10 so only 10 terms in the sum are needed to get the value of 1 n2 within 0.1 of the actual answer. Using the TI89 sum option, we have R10 = 1.54977. Observe that 2 /6 = 1.64493, so it is within 0.1 of the actual answer!
