21
$\begingroup$

The least common multiple of $1,2,\dotsc,n$ is $[1,2,\dotsc,n]$, then

$$\lim_{n\to\infty}\sqrt[n]{[1,2,\dotsc,n]}=e$$


we can show this by prime number theorem, but I don't know how to start

I had learnt that it seems we can find the proposition in G.H. Hardy's number theory book, but I could not find it.

I am really grateful for any help

$\endgroup$
27
$\begingroup$

Let's look how the least common multiple evolves.

If $n > 1$ is a prime power, $n = p^k$ ($k \geqslant 1$), then no number $< n$ is divisible by $p^k$, but $p^{k-1} < n$, so $[1,2,\dotsc,n-1] = p^{k-1}\cdot m$, where $p\nmid m$. Then $[1,2,\dotsc,n] = p^k\cdot m$, since on the one hand, we see that $p^k\cdot m$ is a common multiple of $1,2,\dotsc,n$, and on the other hand, every common multiple of $1,2,\dotsc,n$ must be a multiple of $p^k$ as well as of $m$.

If $n > 1$ is not a prime power, it is divisible by at least two distinct primes, say $p$ is one of them. Let $k$ be the exponent of $p$ in the factorisation of $n$, and $m = n/p^k$. Then $ 1 < p^k < n$ and $1 < m < n$, so $p^k\mid [1,2,\dotsc,n-1]$ and $m\mid [1,2,\dotsc,n-1]$, and since the two are coprime, also $n = p^k\cdot m \mid [1,2,\dotsc,n-1]$, which means that then $[1,2,\dotsc,n] = [1,2,\dotsc,n-1]$.

Taking logarithms, we see that for $n > 1$

$$\begin{align} \Lambda (n) &= \log [1,2,\dotsc,n] - \log [1,2,\dotsc,n-1]\\ &= \begin{cases} \log p &, n = p^k\\ \;\: 0 &, \text{otherwise}.\end{cases} \end{align}$$

$\Lambda$ is the von Mangoldt function, and we see that

$$\log [1,2,\dotsc,n] = \sum_{k\leqslant n} \Lambda(k) = \psi(n),$$

where $\psi$ is known as the second Chebyshev function.

With these observations, it is clear that

$$\lim_{n\to\infty} \sqrt[n]{[1,2,\dotsc,n]} = e\tag{1}$$

is equivalent to

$$\lim_{n\to\infty} \frac{\psi(n)}{n} = 1.\tag{2}$$

It is well-known and easy to see that $(2)$ is equivalent to the Prime Number Theorem (without error bounds)

$$\lim_{x\to\infty} \frac{\pi(x)\log x}{x} = 1.\tag{3}$$

To see the equivalence, we also introduce the first Chebyshev function,

$$\vartheta(x) = \sum_{p\leqslant x} \log p,$$

where the sum extends over the primes not exceeding $x$. We have

$$\vartheta(x) \leqslant \psi(x) = \sum_{n\leqslant x}\Lambda(n) = \sum_{p\leqslant x}\left\lfloor \frac{\log x}{\log p}\right\rfloor\log p \leqslant \sum_{p\leqslant x} \log x = \pi(x)\log x,$$

which shows - the existence of the limits assumed -

$$\lim_{x\to\infty} \frac{\vartheta(x)}{x} \leqslant \lim_{x\to\infty} \frac{\psi(x)}{x} \leqslant \lim_{x\to\infty} \frac{\pi(x)\log x}{x}.$$

For $n \geqslant 3$, we can split the sum at $y = \frac{x}{(\log x)^2}$ and obtain

$$\pi(x) \leqslant \pi(y) + \sum_{y < p \leqslant x} 1 \leqslant \pi(y) + \frac{1}{\log y}\sum_{y < p < x}\log p \leqslant y + \frac{\vartheta(x)}{\log y},$$

whence

$$\frac{\pi(x)\log x}{x} \leqslant \frac{y\log x}{x} + \frac{\log x}{\log y}\frac{\vartheta(x)}{x} = \frac{1}{\log x} + \frac{1}{1 - 2\frac{\log \log x}{\log x}}\frac{\vartheta(x)}{x}.$$

Since $\frac{1}{\log x}\to 0$ and $\frac{\log\log x}{\log x} \to 0$ for $x\to \infty$, it follows that (once again assuming the existence of the limits)

$$\lim_{x\to\infty} \frac{\pi(x)\log x}{x} \leqslant \lim_{x\to\infty} \frac{\vartheta(x)}{x},$$

and the proof of the equivalence of $(1)$ and $(3)$ is complete.

$\endgroup$
  • $\begingroup$ Very nice exposition! +1 $\endgroup$ – Markus Scheuer Feb 10 '16 at 18:53
  • $\begingroup$ It was a pleasure to read this answer. It is truly a textbook quality work. $\endgroup$ – Prism Mar 18 '16 at 0:03
2
$\begingroup$

(Not an answer, just a suggestion.)

It would seem that the power of a prime $p$ in $[1,2,\dots,n]$ is $$\left\lfloor \frac{\log n}{\log p}\right\rfloor.$$ Start of proof: $$\prod_{p\leq n} p^{\frac{1}{n}\left\lfloor \frac{\log n}{\log p}\right\rfloor} = e^{\frac{1}{n}\sum_{p\leq n} \log p \left\lfloor \frac{\log n}{\log p}\right\rfloor}$$

So we need to show that: $$S_n=\frac{1}{n}\sum_{p\leq n} \log p \left\lfloor \frac{\log n}{\log p}\right\rfloor\to 1\text{ as }n\to\infty$$

The crudest estimate of the terms is:

$$\log n - \log p < \log p \left\lfloor \frac{\log n}{\log p}\right\rfloor \leq \log n$$

But that doesn't seem to be good enough. It does show that the limsup is less than $e$.

We can certainly see that $\limsup S_n\leq 1$ from the prime number theorem, since:

$$S_n\leq \frac{1}{n}\sum_{p\leq n}\log n = \frac{\pi(n) n}{\log n}\to 1$$

$\endgroup$
  • 1
    $\begingroup$ Psst! Chebyshev says $\psi$ could be interesting. Von Mangoldt agrees. $\endgroup$ – Daniel Fischer Jun 14 '14 at 19:57
  • $\begingroup$ I remember doing this exercise and struggling with the liminf and limsup too. Can you imagine the frustration? ;-) $\endgroup$ – punctured dusk Jun 14 '14 at 19:58
  • $\begingroup$ As Daniel suggested, the limit follows from $\theta(n)\leqslant\Sigma\cdots\leqslant\pi(n)\log n$, where $\Sigma\log p=\theta(x)\sim\pi(x)\log x\sim x$. (No $\psi$ or $\Lambda$, however.) $\endgroup$ – punctured dusk Jun 14 '14 at 20:13
  • $\begingroup$ @barto $\log [1,2,\dotsc,n] = \psi(n)$. $\endgroup$ – Daniel Fischer Jun 14 '14 at 20:20

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.