A problem on my (last week's) real analysis homework boiled down to proving that, for $a>-1$, $$\sum_{n=1}^\infty\frac{(n-1)!}{n\prod\limits_{i=1}^n(a+i)}=\sum_{k=1}^\infty \frac{1}{(a+k)^2}.$$ Mathematica confirms this is true, but I couldn't even prove the convergence of the original series (the one on the left), much less demonstrate that it equaled this other sum; the ratio test is inconclusive, and the root test and others seem hopeless. It was (and is) quite a frustrating problem. Can someone explain how to go about tackling this?

- 1,987

- 129,973
-
2You can obtain convergence using Raabe's Ratio Test. – xen Oct 25 '11 at 08:51
-
Would going the hypergeometric route be a nuke for you here? – J. M. ain't a mathematician Oct 25 '11 at 09:05
-
@J.M. I'm only vaguely familiar with hypergeometric series, but certainly I'd be interested in whatever approach you have; however, this occurred in a problem in Chapter 2 of Folland's Real Analysis, and I don't believe that we are assumed to have seen hypergeometric series before, so I would hope there is some relatively straightforward way of proving it. – Zev Chonoles Oct 25 '11 at 09:11
-
That's what I was afraid of. This looks amenable to partial fractions, but I expect the algebra to be tedious... – J. M. ain't a mathematician Oct 25 '11 at 09:15
-
@xen: Thanks for the help! We get that $$\lim_{n\to\infty}n\left(\left|\frac{a_{n+1}}{a_n}\right|-1\right)=\lim_{n\to\infty}n\left(\frac{n^2}{(n+1)(a+n+1)}-1\right)=$$ $$\lim_{n\to\infty}\frac{-(2+a)n^2-(1+a)n}{n^2+(2+a)n+(1+a)}=-(2+a)<-1$$ so Raabe's test does the trick. – Zev Chonoles Oct 25 '11 at 09:18
-
@J.M. For the case where $a$ is an integer, using partial fractions is quite literally the same thing as using the beta function trick. It can be a bit tricky to see why both are in fact the same, but trust me, they are! – Eric Naslund Oct 25 '11 at 09:25
-
@Eric: I know, I was trying to disguise the use of nukes right there... ;) – J. M. ain't a mathematician Oct 25 '11 at 09:27
-
@J.M. Until writing it here, I never actually realized the that the beta function trick is likely only to work for a series when partial fractions also works. I found another of my past answers which is precisely that! http://math.stackexchange.com/questions/34351/simpler-way-to-compute-a-definite-integral-without-resorting-to-partial-fraction/34360#34360 – Eric Naslund Oct 25 '11 at 09:51
-
When $a$ is an integer, it can also be done using summation by parts, but in view of Eric's nice general answer, I don't know if I'll bother to type that up (unless somebody is really interested). – Hans Lundmark Oct 25 '11 at 10:09
-
1@Hans: methinks the difference calculus route could be pedagogical; I say go for it! – J. M. ain't a mathematician Oct 25 '11 at 10:21
-
4This was rather a fun question to watch evolving; thanks to everyone! The only thing missing now is a combinatorial proof using objects counted by generating functions given by the summands on the two sides :-) – joriki Oct 25 '11 at 10:46
-
@J.M.: Done. Enjoy! – Hans Lundmark Oct 25 '11 at 11:35
2 Answers
This uses a reliable trick with the Beta function. I say reliable because you can use the beta function and switching of the integral and sum to solve many series very quickly.
First notice that $$\prod_{i=1}^{n}(a+i)=\frac{\Gamma(n+a+1)}{\Gamma(a+1)}.$$ Then
$$\frac{(n-1)!}{\prod_{i=1}^{n}(a+i)}=\frac{\Gamma(n)\Gamma(a+1)}{\Gamma(n+a+1)}=\text{B}(n,a+1)=\int_{0}^{1}(1-x)^{n-1}x{}^{a}dx.$$ Hence, upon switching the order we have that $$\sum_{n=1}^{\infty}\frac{(n-1)!}{n\prod_{i=1}^{n}(a+i)}=\int_{0}^{1}x^{a}\left(\sum_{n=1}^{\infty}\frac{(1-x)^{n-1}}{n}\right)dx.$$ Recognizing the power series, this is $$\int_{0}^{1}x^{a}\frac{-\log x}{1-x}dx.$$ Now, expand the power series for $\frac{1}{1-x}$ to get $$\sum_{m=0}^{\infty}-\int_{0}^{1}x^{a+m}\log xdx.$$ It is not difficult to see that $$-\int_{0}^{1}x^{a+m}\log xdx=\frac{1}{(a+m+1)^{2}},$$ so we conclude that $$\sum_{n=1}^{\infty}\frac{(n-1)!}{n\prod_{i=1}^{n}(a+i)}=\sum_{m=1}^{\infty}\frac{1}{(a+m)^{2}}.$$
Hope that helps,
Remark: To evaluate the earlier integral, notice that $$-\int_{0}^{1}x^{r}\log xdx=\int_{1}^{\infty}x^{-(r+2)}\log xdx=\int_{0}^{\infty}e^{-u(r+1)}udu=\frac{1}{(r+1)^{2}}\int_{0}^{\infty}e^{-u}udu. $$ Alternatively, as Joriki pointed out, you can just use integration by parts.

- 238,052

- 72,099
-
+1, very nice. You can also get that last integral by integration by parts. – joriki Oct 25 '11 at 09:26
-
@Joriki: Hmmm that is a good point! Certainly the easier way. (I was in the Gamma Function mindset) – Eric Naslund Oct 25 '11 at 09:27
-
The only thing I'm wondering now is whether knowledge of the beta function was assumed by Folland when he posed that question as an exercise. (But +1, of course.) – J. M. ain't a mathematician Oct 25 '11 at 09:32
-
5I see my trouble now; the original problem was actually to show that $$\int_0^1x^a(1-x)^{-1}\log(x),dx=\sum_{k=1}^\infty\frac{1}{(k+a)^2}$$ with the instruction to expand as an infinite series and justify switching the order; but I expanded the logarithm, and ended up at the series $$\sum_{n=1}^{\infty}\frac{(n-1)!}{n\prod_{i=1}^{n}(a+i)}$$ with nowhere to go :) Thanks for your excellent answer! – Zev Chonoles Oct 25 '11 at 09:36
-
However, J.M. is correct, the gamma function itself was introduced at the very end of Chapter 2, right before the exercises; no properties of it were developed, nor was the beta function. I see that the identity you used is straightforward to derive (here) but I am not sure if we were supposed to have realized this approach. – Zev Chonoles Oct 25 '11 at 09:37
-
@ZevChonoles: Hmmm, when you did your original expansion, how did you managed evaluate the integrals $$\int_0^1 x^a(1-x)^n,$$ in terms of the products and factorials? I am pretty sure for $a\notin \mathbb{Z}$, evaluating this in terms of the products that you had above is equivalent to showing the famous identity for the Beta Function, $\text{B}(x,y)=\frac{\Gamma(x)\Gamma(y)}{\Gamma(x+y)}$. – Eric Naslund Oct 25 '11 at 09:41
-
2@Eric: Induction on $n$; for $n=0$ we know it's $\frac{1}{a+1}$, then using integration by parts I got $$\int_0^1x^a(x-1)^{k},dx=\left.\frac{x^{a+1}(x-1)^{k}}{a+1}\right|0^1-\frac{k}{a+1}\int_0^1x^{a+1}(x-1)^{k-1},dx=$$ $$(0-0)-\frac{k}{a+1}\left(\int_0^1x^{a+1}(x-1)^{k-1},dx\right)$$ which shows that $$\int_0^1x^a(x-1)^{n},dx=\frac{(-1)^{n}n!}{\prod{i=1}^{n+1}(a+i)}$$ – Zev Chonoles Oct 25 '11 at 09:43
-
1@Zev: oooo right, because $k$ is still an integer! Ok thanks, that is very good! You might also like the proof on Wikipedia for the Beta function/ Gamma function identity. It is basically all the one needs to know about the beta function: (cool integration tricks in the plane) http://en.wikipedia.org/wiki/Beta_function#Relationship_between_gamma_function_and_beta_function – Eric Naslund Oct 25 '11 at 09:47
-
This is quite a bit beyond what was expected of Zev, but I'll leave it here anyway: Mathematica evaluates Zev's starting integral as a Hurwitz zeta function, while both series in the OP evaluate as polygamma functions. The connection between those two functions is well-known to those who know them ;), but the connection would not have been seen without some massaging. – J. M. ain't a mathematician Oct 25 '11 at 10:11
Since at least J. M. asked for it, here's another solution for the case when $a$ is a natural number.
I'll use the forward difference operator $\Delta$, defined by $\Delta f(n) = f(n+1) - f(n)$, and the falling factorial defined by $$ n^{\underline{a}} = \begin{cases} n(n-1)(n-2) \dots (n-a+1), & a > 0, \\ 1, & a=0 \\ \frac{1}{(n+1)(n+2) \dots (n+|a|)}, & a < 0, \end{cases} $$ and satisfying $\Delta n^{\underline{a}} = a n^{\underline{a-1}}$.
The summand, which I'll denote by $F_a(n)$, can be rewritten as $$ F_a(n) = \frac{(n-1)!}{n\prod_{i=1}^n(a+i)} = \frac{(n-1)! a!}{n (a+n)!} = \frac{a!}{n \cdot n(n+1)(n+2) \dots (n+a)} $$ $$= \frac{(a-1)!}{n} \left( -(-a) (n-1)^{\underline{-(a+1)}}\right) = -\frac{(a-1)!}{n} \Delta\left( (n-1)^{\underline{-a}}\right). $$ Using the rule $\Delta(f(n)g(n)) = \Delta f(n) \, g(n+1) + f(n) \Delta g(n)$, we get $$ F_a(n) = - \Delta\left( \frac{(a-1)!}{n} (n-1)^{\underline{-a}}\right) + \Delta\left( \frac{(a-1)!}{n} \right) \, n^{\underline{-a}} $$ $$ = - \Delta\left( \frac{(a-1)!}{n \cdot n (n+1) \dots (n+a-1)} \right) + (a-1)! \left( \frac{1}{n+1} - \frac{1}{n} \right) \frac{1}{(n+1)\dots (n+a)} $$ $$= - \Delta\left( \frac{(a-1)!}{n \cdot n(n+1) \dots (n+a-1)} \right) + F_{a-1}(n+1) - (a-1)! \Delta\left( \frac{(n-1)^{\underline{-a}}}{-a} \right). $$
Summing over $n \ge 1$ gives (because of telescoping in the sums-of-deltas) $$ \sum_{n=1}^{\infty} F_a(n) = \frac{(a-1)!}{1 \cdot a!} + \sum_{n=1}^{\infty} F_{a-1}(n+1) - \frac{(a-1)!}{a} 0^{-\underline{a}} $$ $$ = \frac{1}{a} + \sum_{m=2}^{\infty} F_{a-1}(m) - \frac{1}{a^2} $$ $$ = \sum_{m=1}^{\infty} F_{a-1}(m) - \frac{1}{a^2} $$ (since $F_{a-1}(1) = 1/a$).
Finally, since $F_0(n) = 1/n^2$, we obtain after using this result to work our way down $n$ steps that $$ \sum_{n=1}^{\infty} F_a(n) = \sum_{n=1}^{\infty} F_{a-1}(n) - \frac{1}{a^2} = \dots = \sum_{n=1}^{\infty} F_0(n) - \left( \frac{1}{a^2} + \dots + \frac{1}{1^2} \right) = \sum_{n=a+1}^{\infty} \frac{1}{n^2} = \sum_{k=1}^{\infty} \frac{1}{(k+a)^2}. $$

- 53,395
-
I was expecting you'd be using a as the base and then connect rising and falling factorials, but this is nice! – J. M. ain't a mathematician Oct 26 '11 at 03:33