Prove that $$\lim\limits_{k\to\infty}\left(\frac1k\sum\limits_{n=1}^k\left\lfloor\frac kn\right\rfloor-\ln k\right)=2\gamma-1$$

I have approximated this limit with python:

import math

len = 10000000000

ans = 0.0
x = 1
while (x<len):
    ans += len/x
    x = x + 1

ans = ans/len - math.log(len)

print ans

# this is the value i got for len = 10 billion:   0.154433

If anyone knows anything about this limit I would be extremely interested to hear about it. I came up with it when thinking about the Euler - Mascheroni Constant which is defined similarly.

The main difference here is the I have in a sense rounded the harmonics down to a values that form discrete intervals of [0,1].

It appears that this limit is one less than twice the Euler - Mascheroni Constant or at least very close according to the program I wrote. If anyone could prove this or suggest any ways I might attempt to do so that would be great.

Stated more clearly where L is the limit and γ is the Euler - Mascheroni Constant:

L = 2γ-1
  • $\begingroup$ I guess there is an error in your python code. $\endgroup$ – The Dead Legend May 11 '17 at 5:13
  • $\begingroup$ where is the error? $\endgroup$ – mathew May 11 '17 at 5:21
  • $\begingroup$ Coding. Please wait. $\endgroup$ – The Dead Legend May 11 '17 at 5:21
  • $\begingroup$ It should run as is. It runs on my mac. I have this version of python: 2.7.10 $\endgroup$ – mathew May 11 '17 at 5:24
  • 1
    $\begingroup$ We can write $n^{-1}\sum_{k=1}^n\lfloor n/k\rfloor=H_n-\theta_n$ where $$\theta_n:=\frac1n\sum_{k=1}^n\left\{\frac{n}k\right\}$$ Hence $\theta_n\in (0,1)$ for all $n\in\Bbb N$. Then it seems that $$\lim_{n\to\infty}\frac1n\left(\sum_{k=1}^n\left\lfloor\frac{n}k\right\rfloor\right)-\ln n\in(\gamma-1,\gamma)$$ $\endgroup$ – Masacroso May 11 '17 at 6:34

Note that

$$\tag{1}\frac1k\sum\limits_{n=1}^k\left\lfloor\frac kn\right\rfloor-\ln k = -\frac1k\sum\limits_{n=1}^k\left(\frac{k}{n}-\left\lfloor\frac kn\right\rfloor\right)-\left( \sum_{n=1}^k \frac{1}{n} - \ln k\right).$$

We have

$$\lim_{k \to\infty}\left( \sum_{n=1}^k \frac{1}{n} - \ln k\right) = \gamma,$$

and the first term on the RHS of (1) is a Riemann sum converging to

$$\lim_{k \to\infty}-\frac1k\sum\limits_{n=1}^k\left(\frac{k}{n}-\left\lfloor\frac kn\right\rfloor\right) = - \int_0^1 \left\{\frac{1}{x} \right\}\, dx, $$

where $\{ \cdot \}$ denotes fractional part.

It remains to show that the integral equals $\gamma - 1$.

This follows from

$$\begin{align}\int_0^1 \left\{\frac{1}{x} \right\}\, dx &= \int_1^\infty \frac{\{y\}}{y^2}\, dy \\ &= \sum_{k=1}^\infty \int_k^{k+1} \frac{y-k}{y^2} \, dy \\ &= \sum_{k=1}^\infty \left( \ln \frac{k+1}{k} - \frac{1}{k+1}\right) \\ &= \lim_{n \to \infty} \sum_{k=1}^n \left(\ln \frac{k+1}{k} - \frac{1}{k+1} \right) \\ &= \lim_{n \to \infty} \left(\ln (n+1) - \sum_{k=1}^n \frac{1}{k+1}\right) \\ &= 1 - \gamma \end{align}$$


$$\lim_{k \to \infty} \left(\frac1k\sum\limits_{n=1}^k\left\lfloor\frac kn\right\rfloor-\ln k \right) = \gamma -1 + \gamma = 2 \gamma -1$$

  • $\begingroup$ Really well done, it is clearly explained. :-) $\endgroup$ – user90369 May 11 '17 at 10:07

Well I do not think your Python code has a bug.

To prove it you have to use Dirichlet's asymptotic formula(Theorem 3.3).(You can see Apostol's analytic number theory P$57$).I do not want to tell you how to prove that formula(Because It can be found in the textbook,It is not worth much). I only tell you how to use it(See picture). enter image description here $d(n)$ means divisor function,it equals how many divisor $n$ has.

Now evaluate $S_1= \sum\limits_{n \leq k}d(n)$.Does it equal to $S_2=\sum\limits_{n=1}^k\lfloor\frac{k}{n}\rfloor$?If the answer is Yes,then we can reduce it to the theorem showed in the picture(Theorem $3.3$).The answer is actually yes and proof is easy.For $n=1$,$k$ positive integers no more than $k$ has divisor $1$,for $n=2$,$\lfloor \frac{k}{2} \rfloor$ integers has divisor $2$.for $n=3,4,...,k$,case nearly the same.Then $S_1=S_2$,Dirichlet's asymtotic formula comes to use(We can reduce your problem to the theorem 3.3 in the picture).

Well,the answer shows that your problem has a background of number theory.


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.