18

Let $X_1, X_2,\dots$ be a sequence of strictly increasing positive integers.

For each $n\ge1$, let $W_n$ be the least common multiple of the first $n$ terms $X_1, X_2,\dots, X_n$ of the sequence.

I need to prove the following statement:

The series $1/W_1+1/W_2+\dots+1/W_n\;(n\to\infty)$ is a convergent series.

I tried several ways, including the hint given below, but I have no luck to overcome this problem. I'd like to learn the methods that can be used to prove that such a series converges.

Klangen
  • 5,459
Electro82
  • 1,156

4 Answers4

8

$\newcommand{\lcm}{\operatorname{lcm}}$ Note that for $n>1$ the least common multiple $W_n$ of natural numbers $1\leq X_1<X_2<\ldots<X_n$ fulfills \begin{align*} \lcm(X_{n-1},X_n) \leq W_n \end{align*}

The following theorem was originally conjectured by P. Erdös:

Theorem: Let $X_0,\ldots,X_n$ be integers satifying $1\leq X_0<X_1<\ldots<X_n$ then

\begin{align*} \frac{1}{\lcm(X_0,X_1)}+\frac{1}{\lcm(X_1,X_2)}+\cdots+\frac{1}{\lcm(X_{n-1},X_n)}\leq 1-\frac{1}{2^n}\tag{1} \end{align*}

From (1) we obtain by noting that $W_1=\lcm(1,X_1)$

\begin{align*} \frac{1}{W_1}&+\frac{1}{W_2}+\frac{1}{W_3}+\cdots+\frac{1}{W_n}\\ &\leq \frac{1}{W_1}+\frac{1}{\lcm(X_1,X_2)}+\frac{1}{\lcm(X_2,X_3)}+\cdots+\frac{1}{\lcm(X_{n-1},X_n)}\\ &\leq 1-\frac{1}{2^{n}} \end{align*}

and OPs claim follows.

A proof of the theorem was given by D. Borwein in the paper A sum of reciprocals of least common multiples.

Proof (D. Borwein): For $j=1,2,\ldots,n$ let $$S_j=\frac{1}{\lcm(X_0,X_1)}+\cdots+\frac{1}{\lcm(X_{j-1},X_j)}$$

then $\lcm(X_{j-1},X_j)=u_jX_{j-1}=v_jX_{j}$ where $u_j>v_j\geq 1$. Hence \begin{align*} \frac{1}{\lcm(X_{j-1},X_j)}\leq \frac{1}{X_j}\tag{2} \end{align*} It follows from (2) that \begin{align*} \frac{1}{\lcm(X_{j-1},X_j)}\leq(u_j-v_j)\frac{1}{\lcm(X_{j-1},X_j)}=\frac{1}{X_{j-1}}-\frac{1}{X_j}\tag{3} \end{align*} We obtain from (3) \begin{align*} S_j\leq\frac{1}{X_0}-\frac{1}{X_j}\tag{4} \end{align*}

Now we show (1) is valid by considering all possible conditions on $X_0,X_1,\ldots,X_n$

Case 1: $X_n\leq 2^n$. It follows from (4),

\begin{align*} S_n\leq 1 - \frac{1}{X_n}\leq 1-\frac{1}{2^n} \end{align*}

Case 2: $X_j>2^j$ for $1\leq j \leq n$. We obtain by (2) \begin{align*} S_n\leq\frac{1}{X_1}+\frac{1}{X_2}+\cdots+\frac{1}{X_n}<\frac{1}{2}+\frac{1}{2^2}+\cdots+\frac{1}{2^n}=1-\frac{1}{2^n} \end{align*}

Case 3: $X_k\leq2^k$ for some positive integer $k<n$ and $X_j>2^j$ for $k+1\leq j\leq n$. Then, by (2) and (4) \begin{align*} S_n&=S_k+\frac{1}{\lcm(X_{k},X_{k+1})}+\cdots+\frac{1}{\lcm(X_{n-1},X_n)}\\ &<1-\frac{1}{2^k}+\frac{1}{2^{k+1}}+\cdots+\frac{1}{2^n}\\ &=1-\frac{1}{2^n} \end{align*}

We conclude (1) holds in all cases and the theorem follows.

Markus Scheuer
  • 112,413
  • I'm try to match what you write in "Proof (D. Borwein)", with the problem I posted. When you said lcm(X0,X1) that is my w1 and lcm(x1,x2) is my w2?. In other order of thing, the solution of my problem is a solution or proff of the theorem you posted, no? Thanks again, and I'm sorry to ask silly questions, I'm only don't want have any gaps understanding this problem/topic. – Electro82 Nov 19 '15 at 16:12
  • @Electro82: Note, that $W_1=X_1=lcm(1,X_1)$ and $W_2=lcm(X_1,X_2)$. But in general we have for $n>2$ that $W_n=lcm(X_1,X_2,\ldots,X_n)>lcm(X_{n-1},X_n)$. So, $W_n$ is a product consisting of $n$ factors and it is typically much greater than $lcm(X_{n-1},X_n)$, which are the building blocks in ErdösTheorem. This way we can say, that Erdös Theorem is stronger than your claim. The series in Erdös Theorem is a convergent majorant of your series showing that your series converges. – Markus Scheuer Nov 19 '15 at 22:04
  • 1
    echo. It is a stronger result. WLOG, $X_1>1$ in the original Q, From (4) with $X_0=1$ we have $\sum_{ j \leq n} 1/W_ j \leq S_n\leq 1-1/X_n$, and $X_n$ is strictly increasing in the Q – DanielWainfleet Nov 21 '15 at 20:20
4

Since each of $X_1,\dots,X_n$ are distinct and divide $W_n$ we have that $$n\leq d(W_n)$$ where $d(n)$ is the number of divisors function. Using the classical result of Severin Wigert that $$\limsup_{n\rightarrow\infty}\frac{\log d(n)}{\log n/\log\log n}=\log2,$$ which is obtained by simply considering integers of the form $\prod_{p\leq x}p$, [See Montgomery and Vaughn chapter 2] we have that $n$ sufficiently large, $$\log n\leq\log d(W_{n})\leq\frac{\log W_{n}}{\log\log W_{n}},$$ and so $$W_{n}\geq n^{\log\log n}$$ when $n$ is sufficiently large. This implies that $$\sum_{n}\frac{1}{W_{n}}$$ converges as desired.

Eric Naslund
  • 73,551
  • Does it matter that $W_n$ can be smaller than the product of the first few primes? For example, 24 and 30 both have eight factors. Then $\sum W_n$ would be greater. – Empy2 Nov 17 '15 at 15:29
  • @Michael: In short it doesn't matter in the asymptotics I gave. The bound $$\limsup_{n\rightarrow\infty}\frac{\log d(n)}{\log n/\log\log n}=\log2$$ is obtained when $n=\prod_{p\leq x} p$, but the maximum of $d(n)$ over $n\leq e^x$ does not have $n$ of that form. It's just that when we take logs, the differences are too small to matter in the limit. – Eric Naslund Nov 17 '15 at 16:30
  • The very elementary $d(k) \leqslant 2\sqrt{k}$ gives you $W_n \geqslant n^2/4$. – Daniel Fischer Nov 27 '15 at 19:07
1

Let $W (n)$ be the average order of the numbers $W_n$, i.e.,

$$W (n) = \frac {1} {n} \sum_{k = 1}^{n} W_k.$$

For any $k$ we have $W_{k + 1} = W_k \cdot m_k$ where $m_k$ is the product of primes not present in the factorization of $X_1, X_2, \cdots, X_k$. Note that $m_k$ are squarefree integers. Note also that it may be an empty product, i.e., $m_k = 1$. Then

$$\sum_{k = 1}^{n} W_k = W_1 \cdot \sum_{k = 0}^{n - 1} \prod_{j = 0}^{k} m_j.$$ It is easy to see (and show by induction) that $\prod_{j = 0}^{k} m_j > 2^k$ so we have

$$\sum_{k = 1}^{n} W_k > W_1 \cdot \sum_{k = 0}^{n - 1} 2^k = (2^n - 1) W_1.$$

Hence, $W (n) > \frac {2^n - 1} {n} W_1.$ Consequently, we have

$$\sum_{n = 1}^{\infty} \frac {1} {W (n)} < \frac {1} {W_1} \sum_{n = 1}^{\infty} \frac {n} {2^n - 1} = \frac {2.74403...} {W_1}$$

So the sum of reciprocals of $W (n)$ converges. Then, by Cesaro summation, we see that $$\sum_{n = 1}^{\infty} \frac {1} {W_n}$$ also converges.

  • First of all, thanks for your answer. Because there are 3 answer. I'm looking to know each of one. The only part I can't see rigth a way is when you jump from the equality to the inequality using that the product of n terms is greater than 2^k. Again thanks for your effort. – Electro82 Nov 22 '15 at 03:01
  • You're welcome, sir. Yes I claimed the product of $k$ terms (not $n$) is greater than $2^k$ under the sum. You can induct on $m_k$ to see that. But more intuitively, notice that $m_k$ can be 1, but this will happen less often, and more usually, $m_k$ is the product of (first powers) of some primes and it's natural to assume that on average they are $> 2$ (which means their geometric mean is $> 2$). –  Nov 22 '15 at 08:04
0

Hint:

If $W_{n+1}>W_n$ then $W_{n+1}$ has at least one more prime factor than $W_n$, so $W_{n+1}\ge 2W_n$

ajotatxe
  • 66,849
  • 1
    But one needs to show that $W_{n+1} = W_n$ can't happen too often. Have you a good hint for that? – Daniel Fischer Nov 13 '15 at 21:10
  • Yes, this seems to be the hard part... – ajotatxe Nov 13 '15 at 21:35
  • 4
    If you count divisors, you easily get a lower bound for $W_n$ that suffices. But it should also be possible to bound the number of times $W_k = W_{k+1}$ can happen for $k < n$ to get convergence from that. I haven't yet seen how, unfortunately. – Daniel Fischer Nov 13 '15 at 21:47