It is easy to show that: $$ \sum_{k=1}^n \frac{1}{k} > \ln(n+1), $$ but the Euler–Mascheroni constant is defined as: $$ \gamma = \lim_{n \to \infty} \left( \sum_{k=1}^n \frac{1}{k} - \ln(n) \right). $$

My question is, why was $\gamma$ defined using $\ln(n)$ and not $\ln(n+1$)?

Are the two definitions identical, or does it simply turn out to be more convenient for other applications to define $\gamma$ using $\ln(n)$?

  • 2
    $\begingroup$ Good question, I never thought about that... $\endgroup$ – Troy Woo Aug 27 '14 at 14:28
  • 7
    $\begingroup$ Yes, they are identical (in the limit). Compute $\lim\limits_{n\to\infty} \log(n+1)-\log(n)$. $\endgroup$ – Ian Mateus Aug 27 '14 at 14:28
  • 1
    $\begingroup$ It makes it possible (?) to write $\displaystyle\left(\sum_{k=1}^n-\int_1^n dk\right)\frac 1 k$. ${}\qquad{}$ $\endgroup$ – Michael Hardy Aug 27 '14 at 14:50
  • 3
    $\begingroup$ In fact subtracting $\;\ln\left(n+\frac 12\right)\,$ is the best choice (the error made is near $\dfrac 1{2n}$ for $-\ln(n+0)$ and $-\ln(n+1)$ while near $\dfrac 1{24\,n^2}$ for $+\dfrac 12$). $\endgroup$ – Raymond Manzoni Aug 27 '14 at 14:51
  • 1
    $\begingroup$ @MGA: I considered $\delta(n,a):= \sum_{k=1}^n \frac{1}{k} - \ln(n+a)-\gamma\;$ then $\,\delta(10,0) \approx 0.049167496$ (near $\frac 1{20}$) while $\delta(10,1) \approx -0.046142683$ (near $-\frac 1{20}$) and $\delta(10,1/2) \approx 0.00037733190$ (near $\frac 1{2650}$). Hoping this clarified things, $\endgroup$ – Raymond Manzoni Aug 27 '14 at 18:18

$$ \left( \sum_{k=1}^{n} \dfrac{1}{k} - \ln(n+1) \right) - \left( \sum_{k=1}^{n} \dfrac{1}{k} - \ln(n) \right)=\ln(n)-\ln(n+1)=\ln\left(\frac{n}{n+1}\right)$$

And $$\lim_n \ln\left(\frac{n}{n+1}\right)=\ln 1=0$$

| cite | improve this answer | |
  • $\begingroup$ Thank you, good answer. But there is still something strange for me: In the above, we could use $\ln(n+A)$ for any finite $A\in \mathbb{N}$, and the limit would still evaluate to $0$, but it's certainly not true in general that $\sum_{k=1}^{n} 1/k > \ln(n+A)$. I wonder what I'm missing? $\endgroup$ – MGA Aug 27 '14 at 14:33
  • 4
    $\begingroup$ @MGA While it is not true in general that we have $\sum_{k=1}^{n} \dfrac{1}{k} > \ln(n+A)$ for all $n$, we always have $\sum_{k=1}^{n} \dfrac{1}{k} > \ln(n+A)$ for $n$ large enough. $\endgroup$ – N. S. Aug 27 '14 at 14:37
  • $\begingroup$ Thanks again, I'll try to wrap my head around that. I'm new to analysis, so I'm not use to being very rigorous. $\endgroup$ – MGA Aug 27 '14 at 14:39
  • 1
    $\begingroup$ @MGA To make it simple, it is the same phenomena which happens with lets say $1-\frac{1000000}{n}$. This sequence converges to $1$, even if the first few terms are negative. It eventually becomes positive..... In general, if a sequence converges to a strictly positive number, it means that from some step further all terms are positive (but it can be negative for a while). A finite number of terms doesn't matter for convergence. $\endgroup$ – N. S. Aug 27 '14 at 14:41

As @N.S. pointed out, it does not make any difference in the limit by choosing $\ln(n+A)$ rather than $\ln(n)$.

The use of $\ln(n)$ comes from the convenient integral formula given by $$\gamma = \int_1^\infty \left( \frac{1}{[x]} - \frac1x \right) dx.$$

Here the choice to use $\ln(n)$ is natural.

Moreover, we can view $\gamma$ as the difference in the areas of $1/x$ and $1/[x]$. It can be seen from this viewpoint, by sliding all of the areas under $1/[x]$ and over $1/x$ to the $y$-axis, that $\gamma$ is bounded by $1$. Since all of these areas fit inside of the square that has the origin as well as $(1,1)$ for corners.

| 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.