Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

As stated, I tried to prove the following:

The theorem seems to be very incompletely phrased, since one can obtain non integer sums of the form

$$\frac{1}{4} + \frac{1}{4} + \frac{1}{4} + \frac{1}{7} + \frac{1}{9}$$


$$\frac{1}{2} + \frac{1}{4} + \frac{1}{4} + \frac{1}{{18}} + \frac{1}{{20}}$$

so further detail is needed. Maybe this should be closed as no longer relevant until I come up with a better phrasing and I consider more initial conditions. The big question would be

Given the set $S$ of $n$ integers

$$S=\{x_1,x_2,\dots,x_n\}$$ what are sufficient conditions on $x_1,x_2,\dots,x_n$ so that $$\eta = \sum_{k \leq n }x_k^{-1}$$ is not an integer?

Though I don't know if this is an important/relevant question to be asking.

THEOREM If $x_1,\dots,x_n $ are pairwise coprime, $x_i\neq 1$, let

$$\mu =\sum_{k=1}^n \frac 1 x_k $$

Then $\mu$ can't be an integer.

PROOF By induction on $n$. Asume the theorem is true for $2, \dots, n-1$. I'll analize the case $k=n$.

$(1)$ It is true for $n=2$. If $$(x_1,x_2)=1 \Rightarrow (x_1 x_2,x_1+x_2)=1$$

The proof is simple. We have that $(x_1,x_2)=1$. Let $d \mid x_1+x_2 , d \mid x_1x_2$. Then

$$d\mid x_1(x_1+x_2)-x_1x_2 \Rightarrow d\mid x_1^2$$

$$d\mid x_2(x_1+x_2)-x_1x_2 \Rightarrow d\mid x_2^2$$

So $$d \mid (x_1^2,x_2^2)=(x_1,x_2)=1 \Rightarrow d=1$$

This means $$\frac{1}{x_1}+\frac{1}{x_2}=\frac{x_1+x_2}{x_1 x_2}=\phi$$

is not an integer.

$(2)$ Let

$$\mu = \frac{1}{x_1}+ \frac{1}{x_2}+\cdots+ \frac{1}{x_{n-1}}+ \frac{1}{x_n}$$


$$x_n \mu-1 = x_n\left(\frac{1}{x_1}+ \frac{1}{x_2}+\cdots+ \frac{1}{x_{n-1}}\right) =x_n \omega$$

By hypothesis, $(x_1,\dots,x_{n-1})=1$ so $\omega$ is not an integer. Thus, if $x_n \mu-1$ were an integer, it must be the case:

$$ x_n\left(\frac{1}{x_1}+ \frac{1}{x_2}+\cdots+ \frac{1}{x_{n-1}}\right) =k \text{ ; } k \text{ an integer }$$

$$ x_n \frac{\tau}{x_1 x_2 \cdots x_{n-1}} =k \text{ ; } k \text{ an integer }$$

$\tau$ is the numerator obtained upon taking a common denominator.

But since $\omega$ is not an integer, then it must be the case

$$x_1 x_2 \cdots x_{n-1} \mid x_n$$

which is imposible. Then $x_n \mu -1$ is not an integer. But since $x_n$ and $1$ are, this means $\mu$ isn't an integer, this is,

$$\mu =\sum_{k=1}^n \frac 1 x_k $$

is not an integer. $\blacktriangle$

NOTE The hypothesis that $x_k \neq 1$ is necessary to avoid sums like

$$\frac{1}{1}+\overbrace{\frac{1}{n}+\cdots +\frac{1}{n}}^{n }=1+n\frac{1}{n}=2$$

however, if $(x_1,\dots,x_n)=1$, the sum

$$\nu =\sum_{k=1}^n \frac 1 x_k +1 $$

will clearly not be an integer.

share|cite|improve this question
The notation $(x_1,\dots, x_k)=1$ does not mean that the numbers are pairwise relatively prime. Presumably you want pairwise relatively prime, because of $\frac{1}{2}+\frac{1}{3}+\frac{1}{6}$ and many others. – André Nicolas Jun 13 '12 at 20:20
How did you get $x_1\cdots x_{n-1}\mid x_n$? – anon Jun 13 '12 at 20:25
@AndréNicolas Sure. I'll make that clear. It is good you understood it, though. – Pedro Tamaroff Jun 13 '12 at 20:25
In the step 1. of induction you go by reduction to contradiction. Let $p$ prime which $p|x_1x_2$ and $p|(x_1+x_2)$. Then $p$ divides to $x_1$ or $x_2$. Put $p|x_1$. As $p|(x_1+x_2)$ too, $p|x_2$. But then $p|(x_1,x_2)=1$ (Contradiction). – H. Kabayakawa Jun 13 '12 at 20:26
@anon $\tau$ can't be divided by $x_1 \cdots x_{n-1}$ by hypothesis, so for $k$ to be an integer, one needs $x_1 \cdots x_{n-1} \mid x_n$. – Pedro Tamaroff Jun 13 '12 at 20:26
up vote 4 down vote accepted

Suppose that $(a_i,b_i)=1$ for $1\le i\le n$, and that $(b_i,b_j)=1$ for $1\le i< j\le n$. Then the denominator, in lowest terms, of $$ \sum_{i=1}^n\frac{a_i}{b_i}\tag{1} $$ is $$ \prod_{i=1}^nb_i\tag{2} $$ This follows by induction from the case $n=2$, and that is true because $$ (a_1b_2+a_2b_1,b_1b_2)=1\tag{3} $$ Suppose not, and there is some prime $p$ that divides both $b_1b_2$ and $a_1b_2+a_2b_1$. If $p\,|\,b_1$, then $p\,|\,a_1b_2$, but since $(a_1,b_1)=1$ and $(b_1,b_2)=1$, $p$ can divide neither $a_1$ nor $b_2$. If $p\,|\,b_2$, then $p\,|\,a_2b_1$, but since $(a_2,b_2)=1$ and $(b_1,b_2)=1$, $p$ can divide neither $a_2$ nor $b_1$. Therefore, $(3)$ must be true.

Therefore, since harmonic sums with pairwise coprime denominators fall under this case, no harmonic sum with pairwise coprime denominators (other than the singleton sum $1$) can be an integer.

share|cite|improve this answer

Edit: The comment about $(x_1,\dots,x_k)$ refers to an earlier iteration of the question.

The notation $(x_1,\dots, x_k)=1$ does not mean that the numbers are pairwise relatively prime. Presumably you want pairwise relatively prime, because of $\frac{1}{2}+\frac{1}{3}+\frac{1}{6}$ and many others.

If the $x_i$ are pairwise relatively prime, and greater than $1$, there is an easy non-induction proof. Suppose to the contrary that the sum of the reciprocals is an integer. Let $M$ be the product of all the $x_i$ except one, say $x_n$. Multiply our sum of reciprocals by $M$. We get an integer. But that is impossible, since $x_n$ does not divide $M$, and all the other terms $\frac{M}{x_i}$ are integers.

share|cite|improve this answer
Wouldn't the proof work for, for example, $$\frac{1}{2}+\frac 1 3 +\frac 1 4 +\frac 1 5$$? – Pedro Tamaroff Jun 13 '12 at 20:30
I mostly want to use this to prove the $n$th harmonic number is not an integer. Clearly some of the numbers in $1+\frac 1 2 +\frac 1 3 +\cdots +\frac 1n $ are not coprime, but in a whole, they are. – Pedro Tamaroff Jun 13 '12 at 20:32
The idea of the proof would work for many non-pairwise relatively prime situations. We just have to be more careful about what to multiply by. – André Nicolas Jun 13 '12 at 20:32
For the harmonic number, the standard argument is to consider the largest power of $2$ less than or equal to $n$. – André Nicolas Jun 13 '12 at 20:33
I'm looking to polish this proof. Maybe word it better, be more precise, that's what I need. =) – Pedro Tamaroff Jun 13 '12 at 20:34

As I noted in the comments, even if $x_n\frac{\tau}{x_1\cdots x_{n-1}}$ is an integer and $\frac{\tau}{x_1\cdots x_{n-1}}$ is not, this is not sufficient to establish that $x_1\cdots x_{n-1}\mid x_n$. However we do not need this divisibility to obtain a contradiction.

Here's how I would salvage your argument. Again we have the induction hypothesis. Then:

  • If $x_n\mu $ is not an integer, then neither is $\mu$ and we are done.
  • If $x_n\mu$ is an integer, then writing $\tau=e_{n-2}(x_1,\cdots,x_{n-1})$ we have $$x_n\frac{\tau}{x_1\cdots x_{n-1}}\in \Bbb Z,~~\frac{\tau}{x_1\cdots x_{n-1}}\not\in\Bbb Z ~\implies~ (x_n,x_1\cdots x_{n-1})>1$$ (justify this!) which contradicts pairwise coprimality (justify!), hence $x_n\mu$ is not an integer.
share|cite|improve this answer

$\rm b_i\:$ pair coprime, $\rm\displaystyle \: \sum_i \dfrac{a_i}{b_i} = n\in\mathbb Z\:\Rightarrow\: mod\ b_j\!:\ a_j = b_j\left[n- \sum_{i\,\ne\, j} \dfrac{a_i}{b_i}\right]\!\equiv 0\:\Rightarrow\,\dfrac{a_j}{b_j}\in \mathbb Z,\, \forall\, j$

share|cite|improve this answer
@Peter Be sure to understand why those fractions exist mod $\rm\:b_{\,j},\:$ viz. $\rm\:1/b_{\,i}\:$ exists by $\rm\:(b_{\,i},b_{\,j})=1.\quad$ – Bill Dubuque Jun 17 '12 at 1:01
@Peter Numbers coprime to the modulus are invertible, for if $\rm\: (b,m) = 1\:$ then by Bezout $\rm\: j\,b+k\,m = 1,\:$ i.e. $\rm\:mod\ m\!:\ j\,b\equiv 1,\:$ so $\rm\: j \equiv 1/b,\:$ so $\rm\: aj\equiv a/b.\:$ Thus fractions with denominator coprime to the modulus always exist. – Bill Dubuque Jun 17 '12 at 1:22

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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