4

Does anyone know the sum of all triangle numbers? I.e 1+3+6+10+15+21... I've tried everything, but it might help you if I tell you one useful discovery I've made:

I know that the sum of alternating triangle numbers, 1-3+6-10... Is equal to 1/8 and that to change 1+3+6... Into 1-3+6... You would subtract 6+20+42+70... which is every other triangular number (not the hexagonals) multiplied by two.

1/8 plus this value is 1+3+6+10+...

A final note: I tried to split the triangle numbers into hexagonals and that series and then I got the squares of the odd numbers. Using dirichlet lambda functions This gave me 0 but I don't think this could be right. A number of other sums gave me -1/24 and 3/8 but I have no idea

6 Answers6

6

Hint:

$$\sum_{k=1}^n\frac{k(k+1)}2=\frac12\sum_{k=1}^nk^2+\frac12\sum_{k=1}^nk$$

and the sum follows at once if you know

$$\sum_{k=1}^mk^2=\frac{n(n+1)(2n+1)}6$$

The sum of all the triangular numers, i.e. an infinite series, clearly diverges (and this, in this sense, the sum doesn't exist).

Timbuc
  • 34,191
  • Yeah, I was looking for infinite sums so I tried Ramanujan summation by multiplying by 4 and subtracting – user194821 Nov 23 '14 at 13:12
  • 1
    @user194821, one necessary condition for any infinite series $;\sum a_n;$ to converge (exist) is that $;\lim a_n=0;$ . Here this doesn't happen. – Timbuc Nov 23 '14 at 13:15
  • 1
    Note that the question is tagged [tag:divergent-series]. The OP wants a value for $$\sum_{n=1}^\infty \frac{n(n+1)}{2}$$ using a summation method for divergent series. – Daniel Fischer Nov 23 '14 at 13:21
  • I know this series is divergent and I also know that it can have a real, non infinite, answer assigned to it. See https://m.youtube.com/watch?v=w-I6XTVZXww if you want to know how these values can be assigned – user194821 Nov 23 '14 at 13:22
  • @DanielFischer Perhaps so, yet he writes that $;1-3+6-10+\ldots=\frac18;$ , which is obviosly false. He also didn't mention Cesaro or stuff. – Timbuc Nov 23 '14 at 13:22
  • 1-3+6-10+... Is not 1/8? Please explain how – user194821 Nov 23 '14 at 13:24
  • 4
    Why is it obviously false? If you play the game of summing divergent series at all, then $$f(z) = \frac{1}{(1-z)^3} = \frac{1}{2}\left(\frac{d}{dz}\right)^2\frac{1}{1-z} = \frac{1}{2}\left(\frac{d}{dz}\right)^2\sum_{k=0}^\infty z^k = \sum_{n=1}^\infty \frac{n(n+1)}{2}z^{n-1}$$ gives you $$\sum_{n=1}^\infty (-1)^{n-1}T_n = f(-1) = \frac{1}{8}.$$ – Daniel Fischer Nov 23 '14 at 13:32
  • @user194821 Same thing as above: the sequence's limit isn't zero and thus its limits, as it is, doesn't exist: $$a_n:=\frac{(-1)^nn(n+1)}2\rlap{;;;,/}\xrightarrow[n\to\infty]{}0\implies \sum_{n=1}^\infty a_n;;;\text{diverges}$$ – Timbuc Nov 23 '14 at 13:33
  • Again, @DanielFischer: I know we can "make" some divergent series converge, yet that is not what was written in the question...or at least I didn't see it that way. – Timbuc Nov 23 '14 at 13:34
  • 4
    @Timbuc I think the tag [tag:divergent-series] makes it pretty clear what the OP wants. I don't like that game too much either, but it exists. – Daniel Fischer Nov 23 '14 at 13:37
  • @DanielFischer I didn't see it that way but I admit it could be that way. Anway, saying the alternating sum of the triangular numbers is $;1/8;$ without specifying what summation method is used is, imo, at least dangerously misleading. – Timbuc Nov 23 '14 at 13:40
  • It certainly is dangerous. It wouldn't be misleading if there is a theorem stating that all (sufficiently reasonable) summation methods that assign a finite value to it will assign the same value. But yeah, people just saying that $\sum \text{divergent series} = \text{value}$ without specifying the summation method irk me too. – Daniel Fischer Nov 23 '14 at 13:46
  • @DanielFischer, Exactly my point. Thanks for the comments. – Timbuc Nov 23 '14 at 13:54
4

Let $g(n)$ denote the $n$th triangular number. Then $f(n)=\sum \limits _{i=1}^n g(n) = n(n+1)(n+2)/6$. Try proofing this via induction.

slinshady
  • 537
  • 3
    Note that the question is tagged [tag:divergent-series]. The OP wants a value for $$\sum_{n=1}^\infty \frac{n(n+1)}{2}$$ using a summation method for divergent series. – Daniel Fischer Nov 23 '14 at 13:20
4

Here is a start.

The sums of the triangle numbers come out as $$1,4,10,20,35,56$$

Differences between adjacent terms come out as $$3,6,10,15,21$$

Differences between these come out as $$3,4,5,6$$

And then the differences become constant $$1,1,1$$

Now you have to take differences three times to get a constant, so this means your formula will be a cubic in $n$. Also if you take differences $r$ times to get a non-zero constant $c$ then the coefficient of $n^r$ will be $\frac c{r!}$.

Here you expect a cubic which begins $\cfrac {n^3}6$

Mark Bennet
  • 100,194
  • The first six are tetrahedral numbers and I noticed that from the partial sums of 1+2+3+4+5+... Are triangle numbers and the final sum is -1/12 then -1/12 is kinda like the last triangle number. The last tetrahedral number is the answer to the sum in question – user194821 Nov 23 '14 at 13:15
  • 3
    Note that the question is tagged [tag:divergent-series]. The OP wants a value for $$\sum_{n=1}^\infty \frac{n(n+1)}{2}$$ using a summation method for divergent series. – Daniel Fischer Nov 23 '14 at 13:21
  • @DanielFischer Sorry I didn't see the tag – Mark Bennet Nov 23 '14 at 13:30
4

The $r$-th triangular number is $$T_r=\frac {r(r+1)}2=\binom {r+1}2$$ i.e. $1, 3,6, 10, ...$ for $r=1, 2, 3, 4, ...$.

The sum of the first $n$ triangular numbers is $$S_n=\sum_{r=1}^n T_r=\color{blue}{\sum_{r=1}^n \binom {r+1}2=\binom {n+2}3}=\frac {(n+2)(n+1)n}6$$ i.e. $1, 4, 10, 20, ...$ for $n=1, 2, 3, 4...$. This is also known as the $n$-th tetrahedral number.

1

First: the given series is not Cesaro- or Abel-summable.

For the Abel-summability we would make the ansatz $$ \lim_{x \to \;^-1} 1+3x+6x^2+\cdots+\binom{k+2}{2}x^k+ \cdots \underset{\mathcal A}= { 1\over(1-x)^3}\underset{\mathcal A}=+\infty $$ $\qquad \qquad \qquad \qquad \qquad \qquad \qquad $ $\mathcal A$ indicating Abel-summation for divergent series


Second: if one looks for alternatives, then shifting the index might come to mind, such that we might try $$ \begin{array}{} &S_2&= &1+3+6+10+... \\ &&\underset{??}=& 1+2+3+4+... \\ &&&+0+1+3+6+... \\ &&\underset{??}=& S_1 + S_2 \qquad \qquad \qquad && &\text{where }S_1=1+2+3+4+... \\ \implies& 0 &=& S_1 &&& \text{and no information for $S_2$} \end{array} $$ which makes such an idea worthless.


Third: a more meaningful ansatz is surely the following. (This has already been proposed by Daniel Fischer, but I repeat this here to show the nice generalization)

By this we avoid the idea of any shifting of the index at all and formulate

$$ S_2(s) = {1\over 1^s} + {3\over 2^s} + {6\over 3^s} + \cdots + {\binom{k+1}2\over k^s}+\cdots = \sum_{k=1}^\infty {\binom{k+1}2\over k^s} $$ which gives convergent series for all $s>3$ We can then observe (and prove, of course) that for the whole continuous range of convergence $$ \begin{array}{} 2 S_2(s) &=& {2\over 1^s} + {6\over 2^s} + {12\over 3^s} + \cdots \\ &=& {1\over 1^s} + {2\over 2^s} + {3\over 3^s} + \cdots \\ &+& {1^1\over 1^s} + {2^2\over 2^s} + {3^2\over 3^s} + \cdots \\ &=& \sum_{k=1}^\infty { 1\over k^{s-1}} + \sum_{k=1}^\infty { 1\over k^{s-2}} \\ &=& 1 \cdot \zeta(s-1) + 1 \cdot \zeta(s-2) \end{array}$$ It must be taken from elsewhere, that we can extend this summation-expression beyond range of convergence, towards $$ S_2 = \lim_{s \to 0} \frac12 (1 \cdot \zeta(s-1)+ 1 \cdot \zeta(s-2))\underset{\mathcal Z}= \frac12 \cdot( - \frac1{12}+ 0) = -\frac1{24} $$


I found it much interesting, how this could be generalized for series of binomials of higher order.
We get this way $$ \begin{array}{} S_1 &=& \sum_{k=1}^\infty \binom{k}{1} &\underset{\mathcal Z}=&1 \cdot \zeta(-1) \\ S_2 &=& \sum_{k=1}^\infty \binom{k+1}{2}& \underset{\mathcal Z}=\frac1{2!} (&1 \cdot \zeta(-1) + 1 \cdot \zeta(-2) )\\ S_3 &=& \sum_{k=1}^\infty \binom{k+2}{3} &\underset{\mathcal Z}=\frac1{3!} (&2 \cdot \zeta(-1) + 3 \cdot \zeta(-2) + 1 \cdot \zeta(-3) )\\ S_4 &=& \sum_{k=1}^\infty \binom{k+3}{4} &\underset{\mathcal Z}=\frac1{4!} (&6 \cdot \zeta(-1) + 11 \cdot \zeta(-2) + 6 \cdot \zeta(-3) + 1 \cdot \zeta(-4))\\ \vdots & & \vdots \end{array}$$ where your series occurs as $S_2$.

Heuristically I found the coefficients applying multiple regression (I can show this simple procedure if requested), and finally found, that those coefficients at the $\zeta(-k)$s are the unsigned Stirlingnumbers first kind.

Displayed in a matrix (top-left from the infinite array):

picture


  • 1
    Always thought this was close to the definition. It's how i encounterd them at first, only to remember them when thinking about products. It's not uncommon for kids to have wondered during high school, "hey what's the 's sum of natural numbers", and learn playfully about pascals triangle, (or some another question). And similair at some point wondered what gives (n+1)(n+2)(n+3)... as coefficients followed up by can i get a clean n^k by removing only previous products and thuse got the stirling numbers of the second kind. Most only didn't realise there was a name for such set of numbers. – Gerben Nov 19 '19 at 15:02
  • Did you figure the general solution of this for S_n? – Gerben Nov 21 '19 at 11:53
  • @Gerben - hmm, what do you mean? Doesn't that matrix of Stirling numbers give that answer for the general case $S_n$? Or do I not understand your question? – Gottfried Helms Nov 21 '19 at 12:19
  • Yes it does ofcourse give an answer, I was wondering along the lines of using the abel-sum you stated first at $c=-1$ and $ \sum_{i=0}^{s-1} frac{|St(s,i)|n^{s-i)(-1)^n}{(2^{s+1}-1)s!}$. I did something along the lines a while ago to find a numerical expression of the zeta function as summation, but because you now want the full sum of the stirling numbers i wondered if it would turn out to a nicer expression. This was the expression i got using the abel-sum – Gerben Nov 21 '19 at 18:21
  • But when i did that with the abel sum, i Didn't had to bother about a denominator which seems to screw things up, although the denominator is the stirling numbers of the second kind. – Gerben Nov 21 '19 at 18:46
  • @Gerben - uff! Your formula in the other thread... that needs time for consideration. Unfortunately, I'm no more so quick in my brain, so likely I'll postpone that other Q until another day. – Gottfried Helms Nov 21 '19 at 18:46
  • yeah i am not so sure anymore about the formula (because if m>s is it 0 everywhere for integers? Only then it would made sense so i fear i made a typo in the formula itself), but the question still stands there as the formula was only how i got there. I should rederive it :) but the idea is still the same as using the your abel-summation above and canceling it out with the stirling numbers of the second kind. – Gerben Nov 21 '19 at 20:00
0

As stated in the first comment the easiest way is $$\sum_{n=1}^{\infty}n^2/2+n/2=\sum_{n=1}^{\infty}n^2/2+\sum_{n=1}^{\infty} n/2$$

Here's an alternative answer if you somehow do not want to use $\zeta(-1)$, nor the fact that $\zeta(-2)=0$ and don't want to do it by parts, by making it alternating, for d>1: $$g(n)=n^2/2+n/2$$

$$\sum_{n=1}^{\infty}g(n)=\sum_{n=1}^{\infty}f(dn)-f(n)=\sum_{n=1}^{\infty}\sum_{p=1}^{d-1} f(n)e^{ip\pi2n/d}$$

$$f(n)=n^2c+nb$$ $$c=\frac{1}{(d^3-1)2}$$ $$b=\frac{1}{(d^2-1)2}$$

Say you want d=2, $$f(n)=n^2/14+n/6$$ $$\sum_{n=1}^{\infty}(n^2/14+n/6)*(-1)^n$$ Which is ofcourse Cesaro/abel sumable or whatever trick you want to apply.

Gerben
  • 593