d’Alembert’s test fails whenever vn+1/vn→1. But “tends to 1” can happen in very different ways, and the way the ratio approaches 1 carries the information we need. §183 measures it against the ratio for ∑n−s, whose behaviour we now know exactly. The result decides series such as the hypergeometric series of Gauss, which contains the binomial series and many others as special cases, and whose terms are products too complicated for the integral test.
The idea
For un=n−s the ratio of successive terms is
unun+1=(1+n1)−s=1−ns+O(n21)
by Taylor’s theorem. So the ∑n−s are exactly the series whose ratio is “1−s/n, up to a much smaller error”, and among them the dividing line is s=1. Now take any series whose ratio has the form 1−a/n+O(1/n2). If a>1, pick s between 1 and a: the ratio of our series is eventually smaller than the ratio of the convergent series ∑n−s, so its terms shrink faster, and it converges. If a<1, the same comparison with a divergent ∑n−s gives divergence.
To read off a from numbers, look at n(1−vn+1/vn), which tends to a.
ExploreReading a from the ratios
Try this.Start with α=0.5, β=1.5, γ=2.5. Watch n(1−vn+1/vn) settle on a, then lower γ until the dashed line crosses the red line at 1. Switch to the partial sums on each side of the crossing.
1+1⋅γαβ+1⋅2⋅γ(γ+1)α(α+1)β(β+1)+⋯
nn(1 − vₙ₊₁/vₙ)
n(1−vn+1/vn)
its limit a=γ+1−α−β
the critical value 1
Show
a=γ+1−α−β=1.5. Since a>1, that is γ>α+β, the series converges: it is eventually dominated by ∑n−s for any s between 1 and a. (Term v300=0.000162.)
What to notice
The quantity n(1−vn+1/vn) settles on a=γ+1−α−β, so the series converges exactly when γ>α+β. For a a little below 1 the partial sums still look as if they might level off; they do not, just as ∑n−0.9 does not. At a=1 exactly the test is silent.
Precise statements
Recall that O(1/n2) stands for a quantity whose absolute value is at most K/n2 for some constant K and all large n (§89).
TheoremA finer ratio testHardy: §183
Let vn>0 and
vnvn+1=1−na+O(n21).(1)
Then ∑vn converges if a>1 and diverges if a<1. Moreover, if a>0 then vn→0.
The case a=1 is left by Hardy to Chapter IX (Examples XC, 5), where comparison with ∑1/(nlogn) shows that the series then diverges.
where α, β, γ are real and none of them is zero or a negative integer.
CorollaryConvergence of the hypergeometric seriesHardy: §183
The series (2) converges if γ>α+β and diverges if γ<α+β. Its terms tend to 0 if γ>α+β−1. In particular the series
1+1m+1⋅2m(m+1)+⋯
converges if m<0 and diverges if m>0.
The last series is (2) with α=m, β=γ=1; it is the binomial series of (1−x)−m at x=1.
Proofs
ProofThe finer ratio test
Strategy
Compare the ratio of ∑vn with the ratio of ∑n−s for a well-chosen s, and use the comparison of ratios from §175: if vn+1/vn≤un+1/un for large n and ∑un converges, so does ∑vn.
The ratio for n−s
By Taylor’s theorem with remainder (§151), applied to (1+h)−s with h=1/n,
(1+n1)−s=1−ns+2n2s(s+1)(1+nθ)−s−2,0<θ<1,
and the last term is O(1/n2) for s>0. So un+1/un=1−s/n+O(1/n2).
a>1: choose 1<s<a
Subtracting,
unun+1−vnvn+1=na−s+O(n21).
Since a−s>0, the first term beats the error for large n: the error is at most K/n2, which is less than (a−s)/n once n>K/(a−s). So vn+1/vn<un+1/un for large n.
Conclude convergence
∑n−s converges because s>1, so by the comparison of ratios ∑vn converges.
a<1: choose a<s<1
Now a−s<0 and the same computation gives vn+1/vn>un+1/un for large n. Since ∑n−s diverges for s<1, the divergence half of the comparison of ratios shows that ∑vn diverges. (If a≤0, take any s with 0<s<1 and s>a: the same argument works.)
vn→0 when a>0
Choose 0<s<a. As above vn+1/vn<un+1/un for n≥n0, and the telescoping argument of §175 gives vn≤Kn−s for n≥n0. Hence vn→0.
∎
ProofThe hypergeometric series
Strategy
Compute the ratio of successive terms, expand it in powers of 1/n, and read off a.
The terms are eventually of one sign
The factors α+n, β+n, γ+n are positive once n exceeds −α, −β and −γ, and none of them is ever zero. So from some point on all terms have the same sign, and changing the sign of the whole series if necessary we may apply the test to positive terms. (Finitely many early terms do not matter.)
The ratio
Writing vn for the term containing n factors in each product,
vnvn+1=(n+1)(γ+n)(α+n)(β+n).
Expand
The numerator is n2+(α+β)n+αβ and the denominator n2+(γ+1)n+γ. Their difference is (γ+1−α−β)n+(γ−αβ), so
Here a=γ+1−α−β. The series converges if a>1, that is γ>α+β, and diverges if γ<α+β; its terms tend to 0 if a>0, that is γ>α+β−1. For the special series, a=1+1−m−1=1−m, which exceeds 1 exactly when m<0.
∎
Worked examples and exercises
ExerciseProducts of odd and even numbers●●○
Let cn=2⋅4⋅6⋯(2n)1⋅3⋅5⋯(2n−1). Show that ∑cn diverges, that ∑cn3 converges, and that ∑cn2 is a borderline case of the test.
Hint
cn+1/cn=(2n+1)/(2n+2). Write it as 1−2n+21 and expand.
Hint
The ratio for cnp is the p-th power of the ratio for cn, and (1−x)p=1−px+O(x2).
Solution
cncn+1=2n+22n+1=1−2(n+1)1=1−2n1+O(n21), so a=21<1 and ∑cn diverges (though cn→0, since a>0).
For cnp the ratio is (1−2n1+O(n−2))p=1−2np+O(n21), so a=p/2. For p=3, a=23>1: convergent. For p=2, a=1: the test is silent; Chapter IX shows divergence. (In fact cn∼1/πn, so ∑cnp converges exactly when p>2.)
ExerciseFactorials over rising products●●○
Let α>0. Show that n≥1∑(α+1)(α+2)⋯(α+n)n! converges if α>1 and diverges if α<1. What happens when α=1?
Hint
The ratio of successive terms is (n+1)/(α+n+1).
Solution
vnvn+1=n+1+αn+1=1−n+1+αα=1−nα+O(n21), since nα−n+1+αα=n(n+1+α)α(1+α). So a=α: convergence for α>1, divergence for α<1. When α=1 the term is n!/(n+1)!=1/(n+1), and the series is the harmonic series without its first term: divergent, as Chapter IX’s result for a=1 predicts.
ExerciseTwo familiar series●○○Hardy: §183
For vn=1/n2 and vn=1/(n(n+1)), compute a in (1), and check that the test gives the right answer.
Solution
n−2(n+1)−2=(1+n1)−2=1−n2+O(n21), so a=2. And 1/(n(n+1))1/((n+1)(n+2))=n+2n=1−n+22=1−n2+O(n21), so again a=2. Both converge. The finer test sees the difference between 1−2/n and 1−1/n (the harmonic series), which d’Alembert’s test, looking only at the limit 1, cannot.
ExerciseThe binomial series at x=1●●○Hardy: §183
For which real m (not 0 or a negative integer) does 1+m+2!m(m+1)+3!m(m+1)(m+2)+⋯ converge? Check your answer for m=−21 by writing out the first terms.
Solution
This is (2) with α=m, β=γ=1, so a=1−m: convergence for m<0, divergence for m>0 (for m>0 every term is at least 1 anyway). For m=−21 the terms are 1,−21,−81,−161,−1285,…: after the first term they all have the same sign, as the proof requires, and their size decreases like n−3/2. The sum is (1−1)1/2=0, which is the binomial theorem at the edge of its range (§222).
Check yourself
Check yourselfChoose one
A positive series has vn+1/vn=1−n2+O(n21). What follows?
Why
Here a=2>1, so the series converges; it behaves roughly like ∑n−2. d’Alembert’s test does fail (the ratio tends to 1), which is exactly why the finer test is needed.
Check yourselfTrue or false?
When a=1 in (1), the test of §183 shows that the series diverges.
Why
False, as far as §183 goes: the argument needs an s strictly between a and 1, which does not exist when a=1. The divergence in that case is proved in Ch. IX by comparing with ∑1/(nlogn), a series that diverges more slowly than any ∑n−s with s<1.
Check yourselfEnter a number
For the hypergeometric series with α=1, β=2, γ=3.5, what is a in (1)?
Why
a=γ+1−α−β=3.5+1−1−2=1.5>1, so the series converges. Equivalently, γ=3.5>α+β=3.
Check yourselfTrue or false?
If γ=α+β−21, the terms of the hypergeometric series tend to 0 but the series diverges.
Why
True. Then a=21: since 0<a<1, the terms tend to 0 (they behave like n−1/2) and the series diverges.
Hardy’s language
Summary
Summary
The ratio for ∑n−s is 1−s/n+O(1/n2). Comparing ratios with it gives a test that sees how the ratio tends to 1.
If vn+1/vn=1−a/n+O(1/n2): convergence for a>1, divergence for a<1; a=1 needs Ch. IX (divergence). If a>0 the terms tend to 0.
In practice: compute n(1−vn+1/vn) and find its limit a.
The hypergeometric series has a=γ+1−α−β: it converges exactly when γ>α+β.