10.5: Alternating Series
Up to this point, we have focused on series whose terms were all positive. For example, in Section 10.3 we learned the Integral Test, which applies to a series with positive terms. Then in Section 10.4, we introduced a pair of tests that permits us to compare two series whose terms are all positive. But what if some terms are negative? In this section we define alternating series, introduce a test to establish whether they converge, and learn to bound the errors in their partial sums. We discuss the following topics:
Alternating Series
Alternating series are series whose terms alternate in sign. Consider the following examples of alternating series: \begin{align} -1 + 2 - 3 + 4 - 5 + \cdots &= \sum_{n = 1}^\infty (-1)^n \, n \cma \label{eq:alt-n} \nl 1 - \frac{1}{2} + \frac{1}{3} - \frac{1}{4} + \frac{1}{5} - \frac{1}{6} + \cdots &= \sum_{n = 1}^\infty \frac{(-1)^{n + 1}}{n} \pd \label{eq:alt-harmonic} \end{align} In general, an alternating series takes the form \[\sum_{n = 1}^\infty (-1)^{n + 1} \, b_n \or \sum_{n = 1}^\infty (-1)^n \, b_n \cma\] where \(b_n \gt 0.\) We call the term \((-1)^n\) or \((-1)^{n + 1}\) the alternator because it changes sign depending on whether \(n\) is odd or even. \(\eqrefer{eq:alt-harmonic}\) is the Alternating Harmonic series since it's the sum of reciprocals whose signs are alternating.
Let's now discuss a test to determine whether an alternating series converges. The Alternating Series Test asserts that \(\sum_{n = 1}^\infty (-1)^n \, b_n\) (where \(b_n\) is positive) converges if \(\lim_{n \to \infty} b_n = 0\) and \(\{b_n\}\) is a decreasing sequence (that is, \(b_{n + 1} \leq b_n\)). In words, the test states that an alternating series converges if its terms are shrinking in magnitude to \(0.\)
The Alternating Series Test can only establish convergence; it cannot prove that a series diverges. If a series does not meet the conditions of the Alternating Series Test, then the test is inconclusive. In many situations, if \(\lim_{n \to \infty} b_n \ne 0,\) then we can use the Divergence Test to prove that \(\sum (-1)^n \, b_n\) diverges.
Figure 1 enables us to grasp the intuition behind the
Alternating Series Test by using a number line.
Suppose that the infinite series \(\sum_{n = 1}^\infty (-1)^{n + 1} \, b_n\)
(where \(b_n\) is positive) converges to \(S.\)
Let \(S_N\) denote the
PROOF
Let's split the proof into two cases—one for even \(N\) and one for odd \(N.\)
Since \(b_n\) is decreasing, we have \(b_2 \geq b_3 \geq b_4\) \(\geq b_5 \geq \dots \geq b_{2N}\) and so
\[
\ba
S_2 &= b_1 - b_2 \geq 0 \cma \nl
S_4 &= S_2 + (b_3 - b_4) \geq S_2 \cma \nl
& \vdotss \nl
S_{2N} &= S_{2N - 2} + (b_{2N - 1} - b_{2N}) \geq S_{2N - 2} \pd
\ea
\]
But observe that
\[S_{2N} = b_1 - (b_2 - b_3) - (b_4 - b_5) - \cdots - (b_{2N - 2} - b_{2N - 1}) - b_{2N} \pd\]
In each set of parentheses, the difference is positive
It turns out that the Alternating Harmonic series, as in \(\eqref{eq:alt-harmonic},\) satisfies \[\lim_{n \to \infty} \frac{1}{n} = 0 \and \frac{1}{n + 1} \leq \frac{1}{n} \pd\] In words, the terms of the Alternating Harmonic series are shrinking in magnitude toward \(0.\) Thus, the Alternating Test asserts that the series converges—unlike the standard Harmonic series, which diverges.
We identify \(b_n = \sqrt{n}/(n + 2),\) but it isn't clear whether this function is decreasing. Let's therefore compare \(b_n\) to the function \(f(x) = \sqrt x/(x + 2),\) which is continuous for all nonnegative \(x.\) Differentiating shows \[ f'(x) = \frac{\dfrac{x + 2}{2 \sqrt x} - \sqrt x}{(x + 2)^2} = \frac{-x + 2}{2 (x + 2)^2 \sqrt x} \cma \] which is negative for all \(x \gt 2.\) Hence, \(f\) is decreasing on \([2, \infty),\) meaning \(b_{n + 1} \leq b_n\) for all \(n \geq 2.\) Since \(\lim_{n \to \infty} b_n = 0,\) the Alternating Series Test states that \[\sum_{n = 1}^\infty (-1)^{n + 1} \frac{\sqrt{n}}{n + 2} \convBoxed\]
REMARK Note that \(b_n\) is decreasing for \(n \geq 2,\) but the series' index begins at \(n = 1.\) Can we still use the Alternating Series Test? Yes; we only require that the terms of \(\{b_n\}\) eventually be decreasing to \(0.\) Writing out some terms shows \[\sum_{n = 1}^\infty (-1)^{n + 1} \frac{\sqrt{n}}{n + 2} = \underbrace{\frac{1}{3}}_{0.333 \dots} - \underbrace{\frac{\sqrt 2 }{4}}_{0.354 \dots} + \underbrace{\frac{\sqrt 3 }{5}}_{0.346 \dots} - \underbrace{\frac{1}{3}}_{0.333\dots} + \underbrace{\frac{\sqrt 5}{7}}_{0.319 \dots} + \cdots \pd \] After the second term \(-\sqrt 2/4\), each incremental term is decreasing in magnitude to \(0.\) Because \(\sum_{n = 2}^\infty (-1)^{n + 1} \, b_n\) converges, \(\sum_{n = 1}^\infty (-1)^{n + 1} \, b_n\) also converges. (Adding or subtracting a finite number of numbers to a series doesn't alter whether it converges or diverges.)
Alternating Series Error Bound
Consider the infinite, convergent alternating series \[S = \sum_{n = 1}^\infty (-1)^n \, b_n \cmaa b_n \gt 0 \pd\] In most cases, it's impossible to evaluate \(S\) explicitly. Instead, we can estimate \(S\) by using a partial sum \(S_N = \sum_{i = 1}^{N} (-1)^i \, b_i.\) Yet how do we know how accurate our partial sum is? An error bound is the maximum amount by which \(S_N\) is inaccurate in estimating the true sum \(S.\) The Alternating Series Error Bound enables us to bound an error using the absolute value of the first omitted term, \(b_{N + 1}.\) In other words, the value of \(S_N\) differs from the value of \(S\) by no more than \(b_{N + 1}.\) Mathematically, we express this fact as \[\underbrace{\left|S - S_N\right|}_{\textrm{magnitude of error}} \leq \underbrace{b_{N + 1}}_{\abs{\textrm{next term}}}. \] But this fact only applies to alternating series; it is not true for series in general.
PROOF The error \(R_N\) is given by \(S - S_N.\) We subtract terms of the partial sum \(S_N\) from the terms of the infinite series \(S \col\) \[ \ba R_N &= S - S_N \nl &= \sum_{n = 1}^\infty (-1)^n \, b_n - \sum_{n = 1}^N (-1)^n \, b_n \nl &= (-1)^{N + 1} \, b_{N + 1} + (-1)^{N + 2} \, b_{N + 2} + (-1)^{N + 3} \, b_{N + 3} + (-1)^{N + 4} \, b_{N + 4} + (-1)^{N + 5} \, b_{N + 5} + \cdots \nl &= (-1)^{N + 1} \par{b_{N + 1} - b_{N + 2} + b_{N + 3} - b_{N + 4} + b_{N + 5} - \cdots} \pd \ea \] Hence, we see \[ \ba \abs{R_N} &= \abs{(-1)^{N + 1} \par{b_{N + 1} - b_{N + 2} + b_{N + 3} - b_{N + 4} + b_{N + 5} - \cdots}} \nl &= b_{N + 1} - b_{N + 2} + b_{N + 3} - b_{N + 4} + b_{N + 5} - \cdots \nl &= b_{N + 1} - \par{b_{N + 2} - b_{N + 3}} - \par{b_{N + 4} - b_{N + 5}} - \cdots \cma \ea \] where each difference in parentheses is positive (or \(0\)) because \(b_{N + 2} \geq b_{N + 3}\) \(\geq b_{N + 4} \geq \cdots.\) Thus, \(\abs{R_N} \leq b_{N + 1}.\) So we conclude that the error in \(S_N\) has a magnitude no greater than \(b_{N + 1}.\) \[\qedproof\]
Alternating Series An alternating series is a series whose terms alternate in sign. In general, an alternating series takes the form \[\sum_{n = 1}^\infty (-1)^{n + 1} \,b_n \or \sum_{n = 1}^\infty (-1)^n \, b_n \cma\] where \(b_n \gt 0.\) The factor \((-1)^n\) or \((-1)^{n + 1}\) is called an alternator since its sign changes based on whether \(n\) is odd or even. The Alternating Series Test states that \[\sum_{n = 1}^\infty (-1)^n \, b_n \cmaa b_n \gt 0 \cma\] converges if \(\lim_{n \to \infty} b_n = 0\) and \(b_{n + 1} \leq b_n\) for all \(n \geq N\) for some nonnegative integer \(N.\) In words, an alternating series converges if its terms are decreasing in magnitude to \(0.\) But this test cannot prove divergence; usually, we use the Divergence Test to assert that an alternating series is divergent. The Alternating Harmonic series is the sum of alternating reciprocals: \begin{equation} 1 - \frac{1}{2} + \frac{1}{3} - \frac{1}{4} + \frac{1}{5} - \frac{1}{6} + \cdots = \sum_{n = 1}^\infty \frac{(-1)^{n + 1}}{n} \cma \eqlabel{eq:alt-harmonic} \end{equation} which converges by the Alternating Series Test.
Alternating Series Error Bound We use the Alternating Series Error Bound to calculate the maximum amount (maximum error) by which a partial sum \(S_N\) differs from the infinite sum \(S.\) Consider the convergent alternating series \[S = \sum_{n = 1}^\infty (-1)^n \, b_n \cmaa b_n \gt 0 \pd\] If \(S_N = \sum_{i = 1}^N (-1)^i \, b_i\) is used to estimate \(S,\) then the Alternating Series Error Bound gives \[\abs{S - S_N} \leq b_{N + 1} \pd\] In words, the magnitude of the error is no greater than the first omitted term in the approximation.