I want to know the set of integer solutions to the following sum: $$\sum\limits_{n=1}^{k}\frac{n}{a_n}=1$$ For instance, the series $a_n=\frac{1}{nk}$ satisfies this for all integer $k$. So far, this is the only solution that I have come up with for a general $k$. Here are some solutions that I have for specific values of $k$:

If $k=2$, $a_1=a_2=\frac{1}{3}$ works.

If $k=3$, $a_1=a_2=\frac{1}{4}, a_3=\frac{1}{12}$ works.

If $k=4$, $a_1=a_2=\frac{1}{5}, a_3=\frac{1}{15}, a_4=\frac{1}{20}$ works.

Are there any generalizations or ways to generate new solutions?

  • 4
    $\begingroup$ You've confused $a_n$ and $1/a_n$ (and confused me through it, too). Your solution is obviously $a_n = nk$, not $a_n = \frac{1}{nk}$. $\endgroup$ – celtschk Nov 21 '17 at 19:40

Yet another generic solution: since $$\sum_{n=1}^{k-2} \frac{n}{2^{n+1}}=1-\frac{k}{2^{k-1}},$$ we can write the fraction $\dfrac{k}{2^{k-1}}$ in the form $$\frac{k}{2^{k-1}}=\frac{k-1}{2^{k-1}}+\frac{1}{2^{k-1}} = \color{red}{\frac{k-1}{2^{k-1}}}+\color{darkviolet}{\frac{k}{k\cdot 2^{k-1}}}.$$ Then for given $k$ ($k\ge 3$) we have solution $$(a_1,a_2,\ldots,a_{k-2},a_{k-1},a_k) = \left(4,8,\ldots, 2^{k-1}, \color{red}{2^{k-1}}, \color{darkviolet}{k\cdot 2^{k-1}}\right).$$

$k=3$: $(4,\color{red}{4},\color{darkviolet}{12})$;
$k=4$: $(4,8,\color{red}{8},\color{darkviolet}{32})$;
$k=5$: $(4,8,16,\color{red}{16},\color{darkviolet}{80})$;
$k=6$: $(4,8,16,32,\color{red}{32},\color{darkviolet}{192})$;
$\cdots$ .

In general, the set of solutions is rather wide:


| cite | improve this answer | |

Another generic solution is to choose $a_1=a_2=\ldots=a_k=\frac{k(k+1)}{2}$ Then you have $$\sum_{n=1}^{k}\frac{n}{a_n} = \frac{2}{k(k+1)}\sum_{n=1}^k n = \frac{2}{k(k+1)}\cdot\frac{k(k+1)}{2} = 1$$

For $k=2$ this gives the solution you've already found; $a_1=a_2=3$.

For $k=3$ it gives $a_1=a_2=a_3=6$.

For $k=4$ it gives $a_1 = a_2 = a_3 = a_4 = 10$.

| cite | improve this answer | |

Let's try the case $k=3$. Write the equation $1/a_1 + 2/a_2 + 3/a_3 = 1$ as

$$ a_3 = \frac{3 a_1 a_2}{a_1 a_2 - 2 a_1 - a_2}$$ This will certainly be an integer if the denominator is $\pm 1$ or $\pm 3$. Let's try for $+1$. So our new equation is $$ a_1 a_2 - 2 a_1 - a_2 = 1 $$ which we can solve for $a_2$: $$ a_2 = \frac{2 a_1 + 1}{a_1 - 1}$$ If $t = a_1 - 1$, this is $(2t+3)/t = 2 + 3/t$. It will be an integer if $t = -3, -1, 1$ or $3$, corresponding to $(a_1,a_2,a_3) = (-2,1,-6), (0,-1,0), (2,5,30)$ and $(4,3,36)$ respectively. Of course $(0,-1,0)$ has to be excluded because it leads to division by $0$ in the original equation.

Similarly, denominator $-3$ leads to solution $(2,1,-2)$, $-1$ to $(2,3,-18)$, and $3$ to $(-4,1,-4)$, $(2,7,14)$ and $(6,3,18)$.

The method can be extended to arbitrary $k$. First solve for $a_k$:

$$a_k = \frac{ k a_1 a_2 \ldots a_{k-1}}{P(a_1, \ldots, a_{k-1})} $$ where $P$ is a polynomial of degree $1$ in each variable. Choose one of the divisors $d$ of $k$, and consider the equation $P(a_1, \ldots, a_{k-1}) = d$, which can then be solved for $a_{k-1}$, again as a rational function of $a_1, \ldots, a_{k-2}$ with denominator of degree $1$ in each of these. Choose a value for the denominator that guarantees $a_{k-1}$ is an integer ($\pm$ a divisor of the content of the numerator). Continue until you get to something of the form $a_2 = \frac{\alpha a_1 + \beta}{a_1 -1}$, where you can take $a_1 = t + 1$ for $t$ a divisor of $\alpha + \beta$.

Taking denominators always $+1$, we get the following solutions:

$$ \eqalign{ k=2: & (2,4)\cr k=3: & (2,5,30)\cr k=4: & (2,5,31,1240)\cr k=5: & (2,5,31,1241,1923550)\cr k=6: & (2,5,31,1241,1923551, 4440055831260)\cr etc.}$$

The sequence $2,5,31,1241,1923551, \ldots$ doesn't seem to be in the OEIS. Maybe it will be soon.

EDIT: Now it is: A295391.

| cite | improve this answer | |

Unless you want to restrict to positive integer solutions, you also have solutions such as




and the obvious generalizations for odd and even $k$.

It might be of interest to count the number of solutions (with or without a positivity restriction) for the first few values of $k$, and see if that's in the OEIS. The cases $k=1$ and $2$ are easy, but $k=3$ already seems a little tricky, at least by hand. (I've tried systematically listing the positive solutions, but I keep making mistakes. A computer program might have better luck.)

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.