Alternating Series Test

Theorem

Suppose {an}n=1 is a sequence of non-negative real numbers satisfying

  1. anan+1 for all nZ>0
  2. limnan=0

Then the alternating series

n=1(1)n1an=a1a2+a3a4+

converges.

Note that the same result holds if the sequence is all non-positive, or if the power of 1 is n instead of n1. This just follows from a substitution of the terms in the series with their negatives, and a reindexing of the sum with leading terms excluded.


Lemma

For m3, and a series {an}n=1 which is non-negative and satisfies

  • anan+1 for all nZ>0

we have that

a1a2n=1m(1)n1ana1a2+a3.
Proof

This result follows by grouping the terms in pairs. Namely we first note

n=1m(1)n1an=(a1a2)+n=3m(1)n1an.

Then because anan+1, we have anan+10. Hence we can group the tail of the series into pairs which are strictly non-negative, and then we have potentially one remaining non-negative term. That is we either have

n=1m(1)n1an=(a1a2)+n=3n oddm1(anan+1)

if m1 is odd, or

n=1m(1)n1an=(a1a2)+n=3n oddm1(anan+1)+am

if m1 is even.

Either way, this expression is a1a2 plus non-negative terms. Therefore we can conclude that

a1a2n=1m(1)n1an.

A similar argument follows for the other inequality, where we write

n=1m(1)n1an=(a1a2+a3)+n=4m(1)n1an

and then group in pairs with negative terms. In particular we have

n=1m(1)n1an=(a1a2+a3)+n=4n evenm1(an+an+1)

if m1 is even, and

n=1m(1)n1an=(a1a2+a3)+n=4n evenm1(an+an+1)am

if m1 is odd.

In this case, we have a1a2+a3 plus non-positive terms, and conclude, with the above inequality, that

a1a2n=1m(1)n1ana1a2+a3.

This lemma will now allow us to prove the main result.

Proof

Define the sequence

Sm=n=1m(1)n1an

of partial sums such that we can define the partial sums of an even number of terms and the partial sum of an odd number of terms as S2m and S2m1 respectively (for m1).

With the same grouping in pairs argument from our lemma, it is clear that S2m is monotonic increasing and S2m1 is monotonic decreasing. Thus with the bounds from this lemma and the monotone convergence theorem, we know that

limmS2mandlimmS2m1

both exist.

As such we have

limmS2mlimmS2m1=limm(S2mS2m1)=limma2m=0

and therefore

L=limmS2m=limmS2m1.

Then, the limit of partial sums of the desired series must be equal to the shared limit between the even and odd terms.

In essence, for any ϵ>0 there exists an N0 and N1 such that

n>N0|S2nL|<ϵandn>N1|S2n1L|<ϵ

so setting N=max{N0,N1}, we have that for any ϵ>0

n>N|S2nL|<ϵand|S2n1L|<ϵ

and given both cases cover any term in the partial sum Sn,

limnSn=L.