Skip to main content

Section 8.5 Alternating Series and Conditional Convergence

Before we move too far forward, let’s circle back to a point made in Why Do We Need These Conditions?. In the Integral Test, we required the terms of our series (and the continuous function we connected it with) to be positive. This was really just a mechanism that allowed us to say, in our proof, that the sequence of partial sums was monotonic. When we accumulate more of a positive thing, the total gets bigger. This is half of what we needed for us to employ the Monotone Convergence Theorem. And because this is such a useful tool, we’ll see more of this “positive term series”condition showing up in the tools we use to see if a series converges.
But that makes this a perfect time to stop and ask a hallowed mathematical question: What happens if that property isn’t there? What happens when our series does not only have positive terms?
We definitely have fewer tools to use, since we don’t get anything that relies on applying the Monotone Convergence Theorem to partial sums. So instead, we’ll take a brief detour into something we call Alternating Series (a series whose terms alternate in sign).

Activity 8.5.1. Which is More Likely to Converge?

We’re going to try to think about what might be different when we analyze an alternating series compared to a series with only positive (or non-negative) terms.
Let’s say that \(\{a_k\}_{k=0}^\infty\) is some sequence of positive real numbers. Now let’s consider the two series:
\begin{align*} \sum_{k=0}^\infty a_k \amp \amp \text{vs} \amp \amp \sum_{k=0}^\infty (-1)^k a_k \end{align*}

(a)

Let’s first consider the sequences of terms: \(\{a_k\}\) compared with \(\{(-1)^k a_k\}\text{.}\) Is either of these more or less likely to converge? Does this tell us anything about whether or not the corresponding series converges?
Hint 1.
Try thinking about how we might find \(\displaystyle\lim_{k\to\infty} (-1)^k a_k\text{,}\) especially using the The Squeeze Theorem.
Hint 2.
What does the Divergence Test say? Is either of these sequences more or less likely to converge to 0 (or not)?

(b)

Now let’s think of the partial sums:
\begin{align*} \left\{\sum_{k=0}^n a_k\right\}_{n=0}^\infty \amp \amp \text{vs} \amp \amp \left\{\sum_{k=0}^n (-1)^k a_k\right\}_{n=0}^\infty \end{align*}
Is either of these sequences more or less likely to converge? Does this tell us anything about whether or not the corresponding series converges?
Hint.
It might be helpful to plot these partial sums! Referencing a picture like Figure 8.2.5 and think about how this will change the plot of partial sums for terms that alternate in sign.

(c)

Now make a conjecture about which infinite series is more likely to converge:
\begin{align*} \sum_{k=0}^\infty a_k \amp \amp \text{vs} \amp \amp \sum_{k=0}^\infty (-1)^k a_k \end{align*}
Remember that \(a_k\gt 0\) for \(k=0,1,2,...\text{,}\) so the only differences are the changes in sign.
Hint.
Remember that series convergence is defined by the behavior of the sequence of partial sums!

Subsection Defining Alternating Series, and the Main Result

Ok so hopefully we have convinced ourselves that a series that has terms that alternate in sign might have an “easier”time converging than a series with only positive terms. Let’s start with a definition (so that we can continue to refer to these types of series easily), and then move towards the main result.

Definition 8.5.1. Alternating Series.

An infinite series \(\displaystyle\sum_{k=0}^\infty a_k\) is called an Alternating Series when \(a_k = (-1)^k|a_k|\) or \(a_k = (-1)^{k+1}|a_k|\) for all \(k=0,1,2,...\text{.}\) That is, the sign of the terms alternates:
\begin{equation*} \sum_{k=0}^\infty a_k = |a_0| - |a_1| + |a_2| - + ... \end{equation*}
So this is the type of series we’re thinking of: the terms perfectly alternate in sign.
Now, we will bring into focus one of the main results about alternating series before investigating these types of series (and how they converge) further.
To lead into this new result, let’s remind ourselves of a few things:
  1. Remember the Divergence Test. We, specifically, want to remember that, in general, we don’t know anything about a series if \(\displaystyle \lim_{k\to\infty} a_k=0\text{.}\)
  2. We have typically been looking at infinite series where we impose a further restriction on the terms: we looked at infinite series with only positive terms. We can get a lot more information about these series!
  3. Now we are looking at infinite series with a different kind of structure on their terms: the signs alternate. So it won’t be a surprise when we get to add information about them!
The big idea that we’ll use the extra information about these alternating series (based on how we defined them) to get more information from the limit of the terms. Normally we don’t get any information when \(\displaystyle \lim_{k\to\infty} a_k = 0\text{:}\) the series of those terms could converge or diverge, and we can’t tell! But with this new structure of the terms (the alternating signs), we’ll actually be able to tell something from the limit of the terms being 0.
Let’s look at it! We’re going to think about visualizing the partial sums.
In Figure 8.2.5, we looked at the partial sums of an infinite series and saw how the terms made up the differences between those partial sums. We’re going to think about this same picture, but think about it through the lense of:
  1. an alternating series (with terms that alternate in sign) where...
  2. the terms of the alternating series approach 0 in the limit: \(\displaystyle \lim_{k\to\infty} a_k = 0\text{.}\) Note that this means that the size of the terms must go to 0 in the limit, \(|a_k|\to 0\text{.}\)
Red points on a plot. They are labeled S_0, S_1, S_2, S_3, S_4, S_5, S_6, ... S_n, S_(n+1). The points bounce up and down, where the even indexed points are above. The vertical distance between each point is green and is labeled a_0, a_1, a_2, a_3, a_4, a_5, a_6, ... a_n, a_(n+1). These are getting smaller as we move from point to point.
Figure 8.5.2. Partial sums of an alternating series.
Let’s note a couple of things:
  • All of the “even-indexed”terms (\(a_0\text{,}\) \(a_2\text{,}\) \(a_4\text{,}\) etc.) are positive, while all of the “odd-indexed”terms (\(a_1\text{,}\) \(a_3\text{,}\) \(a_5\text{,}\) etc.) are negative.
  • This means that all of the “even-indexed”partial sums are big, while all of the “odd-indexed partial sums”are small. Our sequence of partial sums bounces up to an even-index and bounces down to an odd-index.
    As long as the size of the terms (the size of the differences between partial sums) is decreasing like we have pictured, then each “next”even-indexed partial sum is a bit smaller than the “previous”even-indexed partial sum. The same thing is true for the odd-indexed partial sums.
  • The terms themselves represent the distance between these successive partial sums: the difference between \(S_n\) and \(S_{n+1}\) is the term \(a_{n+1}\text{.}\)
So, as long as the sizes of the terms are decreasing (or, as long as the distance between partial sums is decreasing consistently), then what happens when the terms (the distance between partial sums) goes to 0?
The even-indexed partial sums and the odd-indexed partial sums approach each other!

Proof.

This proof will follow the discussion before the statement of the theorem. Mostly, we will just fill in some details and provide some further justification for why what we were noticing must be true.
Let’s start with the conditions of the test:
  • We are considering an Alternating Series, \(\displaystyle \sum_{k=0}^\infty a_k\text{.}\) For our purposes, we’ll assume that we have something like
    \begin{equation*} \sum_{k=0}^\infty a_k = |a_0| - |a_1| + |a_2| - |a_3| +-... \end{equation*}
    where the even-indexed terms are the positive ones. This could be flipped and it wouldn’t make a difference.
  • The size of the terms are decreasing. That is, \(|a_k|\lt |a_{k+1}\) for all \(k=0,1,2,...\)
  • The limit \(\displaystyle\lim_{k\to\infty} a_k=0\text{.}\) Note that this also means that \(\displaystyle\lim_{k\to\infty} |a_k|=0\text{.}\)
We’re going to show that, under these conditions, the alternating series we’re considering must converge. The way that we’ll do this is, no surprise, by invoking Theorem 8.1.10 Monotone Convergence Theorem. We’re going to do it by considering the partial sums in halves: the even-indexed ones and the odd-indexed ones.
First, consider the sequence \(\{S_{2n}\}_{n=0}^\infty = \{S_0, S_2, S_4, ...\}\text{.}\) The difference between successive terms in this sequence (successive even-indexed partial sums) is:
\begin{equation*} S_{2n+2} - S_{2n} = -|a_{2n+1}| + |a_{2n+2}|\text{.} \end{equation*}
Since the terms of the alternating series are decreasing in size, we know that \(|a_{2n+2}|\lt |a_{2n+1}|\text{,}\) which means that \(S_{2n+2}-S_{2n}\lt 0\text{,}\) and so \(S_{2n} \gt S_{2n+2}\text{.}\)
All of this which is to say, \(\{S_{2n}\}_{n=0}^\infty\) is a decreasing sequence.
We can apply the same reasoning to the sequence \(\{S_{2n+1}\}_{n=0}^\infty = \{S_1, S_3, S_5, ...\}\text{.}\) We know the differences between successive odd-indexed partial sums is:
\begin{equation*} S_{2n+3} - S_{2n+1} = -|a_{2n+3}| + |a_{2n+2}|\text{.} \end{equation*}
This time, though, \(|a_{2n+2}|\gt |a_{2n+3}|\) and so the difference is positive: \(S_{2n+3} - S_{2n+1}\gt 0\) which means that \(S_{2n+3} \gt S_{2n+1}\text{.}\) The sequence \(\{S_{2n+1}\}_{n=0}^\infty\) is an increasing sequence.
We’re getting close! We have monotonic sequences. Now we just need to show bounds, and then we’ll show that each of these sequences converges. Then, we’ll show that they converge to the same thing.
Getting an upper bound on the odd-indexed partial sums and a lower bound on the even-indexed partial sums is pretty easy. Let’s consider subsequent partial sums, \(S_{2n}\) and \(S_{2n+1}\text{.}\)
\begin{align*} S_{2n+1} - S_{2n} \amp =-|a_{2n+1}|\\ S_{2n+1} - S_{2n} \amp \lt 0\\ S_{2n+1} \amp \lt S_{2n} \end{align*}
Ok so this is easy: we can just pick any odd-indexed partial sum to be the lower bound on the even-indexed partial sums, and vice versa.
So \(S_0\) is an upper bound on \(\{S_{2n+1}\}_{n=0}^\infty\text{,}\) since every other \(S_{2n}\) is less than \(S_0\text{,}\) and all of \(S_{2n+1}\) partial sums are less than \(S_{2n}\text{:}\)
\begin{align*} S_0 \amp \geq S_{2n} \amp \text{for } n=0,1,2,... \\ S_0 \amp \geq S_{2n} \gt S_{2n+1} \amp \text{for } n=0,1,2,... \\ S_0 \amp \gt S_{2n+1} \amp \text{for } n=0,1,2,... \end{align*}
Similarly, we can say the same thing about \(S_1\) being a lower bound for the even-indexed partial sums:
\begin{align*} S_1 \amp \leq S_{2n+1} \amp \text{for } n=0,1,2,... \\ S_1 \amp \leq S_{2n+1} \lt S_{2n} \amp \text{for } n=0,1,2,... \\ S_1 \amp \lt S_{2n} \amp \text{for } n=0,1,2,... \end{align*}
So we have shown that the sequences \(\{S_{2n}\}_{n=0}^\infty\) and \(\{S_{2n+1}\}_{n=0}^\infty\) are both monotonic and bounded and so both of these sequences must converge.
Now we can show that they converge to the same thing.
Since \(\{S_{2n}\}_{n=0}^\infty\) converges, let’s say that there is some number \(S_E\) where
\begin{equation*} \lim_{n\to\infty} S_{2n} = S_E\text{.} \end{equation*}
Similarly, there is a number \(S_O\) where
\begin{equation*} \lim_{n\to\infty} S_{2n+1} = S_O\text{.} \end{equation*}
Now we can use the fact that the limit of the terms is 0:
\begin{align*} \lim_{k\to\infty} a_k \amp=0 \\ \lim_{n\to\infty} a_{2n+1} \amp =0\\ \lim_{n\to\infty} \left(S_{2n+1}-S_{2n}\right) \amp=0 \\ S_O-S_E\amp=0 \end{align*}
So the numbers that these two sequences of partial sums converge to are actually equal to each other.
So, finally, we know that under the conditions we started with, the alternating series must converge.
We can actually get another result really easily from this one! It’s about how we might approximate the value that an infinite series converges to.
We know that if a series converges, it’s because the limit of the partial sums exists. So we “just”need to find what the real number, \(S\text{,}\) is when
\begin{equation*} \lim_{n\to\infty} \sum_{k=0}^n a_k = S\text{.} \end{equation*}
We said earlier, though, that this is pretty hard to do! We can always approximate this limit, though, by looking at a big partial sum: if the limit exists, then probably adding up the first 10,000 terms will be a pretty good approximation (since we don’t expect the partial sums to change, much, as \(n\) gets bigger). But how many terms is enough to give us a good enough approximation?
That’s a hard question to answer, until we think about alternating series.

Activity 8.5.2. Approximating an Alternating Series.

Let’s look, again, at the picture of the partial sums of an alternating series in Figure 8.5.2. We’re going to assume that the series converges, which means that:
  • \(\displaystyle \lim_{n\to\infty} S_n\) exists.
  • \(\displaystyle\lim_{n\to\infty}a_n = 0\text{.}\)
Let’s add to our figure.
Red points on a plot. They are labeled S_0, S_1, S_2, S_3, S_4, S_5, S_6, ... S_n, S_(n+1). The points bounce up and down, where the even indexed points are above. The vertical distance between each point is green and is labeled a_0, a_1, a_2, a_3, a_4, a_5, a_6, ... a_n, a_(n+1). These are getting smaller as we move from point to point. This means that the red points are getting closer to a grey dashed line that cuts through the points labeled lim_(n to infinity) S_n.
Figure 8.5.4.
(a)
Why are the even-indexed partial sums sitting above the odd-indexed partial sums?
Hint.
We started, in this case, with a positive term \(a_0\text{,}\) and then the rest of the signs alternate. How does this change the partial values of partial sums?
(b)
Why are the even-indexed partial sums sitting above the horizontal line, \(\displaystyle \lim_{n\to\infty} S_n\text{?}\)
Hint.
We know that the sizes of the terms, \(|a_n|\text{,}\) decreases. This means that the sequence \(\{S_0, S_2, S_4, ...\}\) is decreasing.
What does that mean about how these partial sums relate to the limit of partial sums?
(c)
Why are the odd-indexed partial sums sitting below the horizontal line, \(\displaystyle \lim_{n\to\infty} S_n\text{?}\)
Hint.
We know that the sizes of the terms, \(|a_n|\text{,}\) decreases. This means that the sequence \(\{S_1, S_3, S_5, ...\}\) is increasing.
What does that mean about how these partial sums relate to the limit of partial sums?
(d)
If we were trying to approximate the value of \(\displaystyle\lim_{n\to\infty} S_n\text{,}\) how can we use the partial sums to build an interval that approximates the value?
Solution.
Since all of the even-indexed partial sums are upper bounds of any approximation and the odd-indexed partial sums are lower bounds.
This means that any pair of partial sums serves as an approximation:
\begin{align*} S_1 \lt \amp\lim_{n\to\infty} S_n \lt S_0 \\ S_{2n+1} \lt \amp \lim_{n\to\infty} S_n \lt S_{2n} \end{align*}

Subsection Convergence, More Carefully

Let’s circle back to an important point from Activity 8.5.1: an alternating series is more likely to converge than its positive-term counterpart.
Let’s look at a classic example of this: the alternating harmonic series.

Activity 8.5.3. The Alternating Harmonic Series Converges.

Let’s consider the alternating harmonic series, as written below:
\begin{equation*} \sum_{k=1}^\infty \frac{(-1)^{k+1}}{k} = 1-\frac{1}{2}+\frac{1}{3}-\frac{1}{4}+\frac{1}{5}-\frac{1}{6}+-... \end{equation*}
(b)
Find \(S_6 = \displaystyle\sum_{k=1}^{k=6} \frac{(-1)^{k+1}}{k}\text{.}\)
(d)
Use technology to find \(S_{1000}\text{.}\) Compare this value to \(\ln(2)\text{.}\) Would it surprise you for someone to claim that the alternating harmonic series converges to \(\ln(2)\text{?}\)
So we have some evidence that says that the alternating harmonic series converges to \(\ln(2)\text{.}\) Let’s look at another “version” of this “same” series. You’ll notice that we’re using scare-quotes on “version” and “same,” and that’s what we’re going to investigate.

Activity 8.5.4. The Alternating Harmonic Series Converges (Again).

Now let’s consider a new, rearrranged, version of the alternating harmonic series:
\begin{equation*} 1-\frac{1}{2}-\frac{1}{4}+\frac{1}{3}-\frac{1}{6}-\frac{1}{8} +\frac{1}{5}-\frac{1}{10}-\frac{1}{12}+ ... \end{equation*}
We can write this in summation notation as:
\begin{equation*} \sum_{k=1}^\infty \left(\frac{1}{2k-1}-\frac{1}{4k-2}-\frac{1}{4k}\right)\text{.} \end{equation*}
(a)
First, confirm that all of the terms from the alternating harmonic series will eventually show up in this series. Convince yourself that we are truly adding (and subtracting) all of the same values with just a different order.
(c)
Add up the first few terms of this series to find the value of a partial sum. You can choose how many terms you add. Does it look like it will also converge to \(\ln(2)\text{?}\)
(d)
Use technology to add up many terms of this series. Can you convince yourself what this series converges to?
This is strange! We have two series that seem to be the same thing (one is just a rearranged version of the other) that both converge, but they converge to different things! One seemingly converges to \(\ln(2)\) while the other converges to half of that: \(\dfrac{\ln(2)}{2}\text{.}\) Very strange!
This doesn’t seem to follow the normal rules of addition: we lose the normal associative property of addition, where the order or the way that we group terms to add typically doesn’t matter. Here it does!
It turns out (and we won’t prove this) that this type of convergence happens only for alternating series, and further it only happens for alternating series whose positive-term counterpart diverges.

Definition 8.5.6. Conditional (and Absolute) Convergence.

If \(\displaystyle \sum_{k=0}^\infty a_k\) is a converging alternating series, then we say that the alternating series \(\displaystyle\sum_{k=0}^\infty a_k\) converges conditionally if \(\displaystyle\sum_{k=0}^\infty |a_k|\) diverges.
If, instead, the positive-term series \(\displaystyle \sum_{k=0}^\infty |a_k|\) converges as well, then we say that the alternating series \(\displaystyle\sum_{k=0}^\infty a_k\) converges absolutely.
So now, for a series whose terms alternate in size, we can:

Practice Problems Practice Problems

1.

What is an “alternating series?” What are some reasons that we need to consider these as a separate class of infinite series?

2.

Explain the differences between the Divergence Test and the Alternating Series Test. What do each tell us about the infinite series? How do we do each of the tests?

3.

Give an example of an alternating series that converges conditionally. Explain your reasoning, and show that it converges conditionally.

4.

Give an example of an alternating series that converges absolutely. Explain your reasoning, and show that it converges absolutely.

5.

Explain how an alternating series could converge conditionally. What about alternating series makes this even possible? Can a positive-term series converge conditionally? Why or why not?

6.

For each of the following alternating series, determine whether the series diverges, converges conditionally, or converges absolutely.
(a)
\(\displaystyle\sum_{k=1}^\infty \left( \frac{(-1)^{k+1} \tan^{-1}(k)}{k^4} \right)\)
(b)
\(\displaystyle\sum_{k=1}^\infty \left( \frac{(-1)^{k+1}}{\sqrt{k^2+3}} \right)\)
(c)
\(\displaystyle\sum_{k=1}^\infty \left( \frac{(-1)^k}{\sqrt[k]{k}} \right)\)
(d)
\(\displaystyle\sum_{k=2}^\infty \left( \frac{(-1)^{k+1}}{\sqrt[3]{k^2-k}} \right)\)
(e)
\(\displaystyle\sum_{k=0}^\infty \left( \frac{(-1)^k}{\sqrt{k^3+1}} \right)\)
(f)
\(\displaystyle\sum_{k=1}^\infty \left( \frac{(-2)^{k+1}}{k^k} \right)\)
(g)
\(\displaystyle\sum_{k=1}^\infty \left( \frac{(-1)^{k+1} k}{k^5-k^2+1} \right)\)
(h)
\(\displaystyle\sum_{k=1}^\infty \left( \frac{\cos(\pi (k-1))}{k^2\ln(k+1)} \right)\)
(i)
\(\displaystyle\sum_{k=0}^\infty \left( \frac{\cos(\pi k)}{e^k + \pi^k} \right)\)

7.

In Section 8.4, Practice Problem 3, you found the values of \(p\) (with \(p\gt0\)) such that the series \(\displaystyle\sum_{k=1}^\infty \frac{1}{k^p}\) converges.
What are the values of \(p\) (with \(p\gt 0\)) such that
\begin{equation*} \displaystyle\sum_{k=1}^\infty \frac{(-1)^{k+1}}{k^p} \end{equation*}
converges conditionally?

8.

In Section 8.4, Practice Problem 4, you found the values of \(r\) (with \(r\gt0\)) such that the series \(\displaystyle\sum_{k=0}^\infty \frac{1}{r^k}\) converges.
Are there any values of \(r\) (with \(r\gt 0\)) such that
\begin{equation*} \displaystyle\sum_{k=0}^\infty \frac{(-1)^{k}}{r^k} \end{equation*}
converges conditionally?