-1
$\begingroup$

Let $\zeta$ denote the Riemann zeta function. In this answer: https://mathoverflow.net/a/314066/133634, @Paul Garret considers the representation

$$\frac{\zeta(s)}{s} = \int_1^\infty (\sum_{1 \le n \le x} 1) x^{-s-1}dx$$ for $\Re(s)>1$, and then claims (without proof) that this representation yields

$$\frac{\zeta(s)}{s}+\frac{\zeta(s)}{1-s}= \int_0^\infty ((\sum_{1 \le n \le x} 1-x)-x(\sum_{1 \le n \le x} \frac1n-\log x-\gamma)) x^{-s-1}dx$$ for $\Re(s)\in(0,1)$, where $\gamma$ is the Euler constant.

But i can't see how the second representation follows from the first. Can anyone please explain ?

$\endgroup$
4
  • $\begingroup$ Actually, it was @reuns who composed that answer. The derivation is pretty standard, in any case. $\endgroup$ Dec 19, 2018 at 22:45
  • $\begingroup$ @Paul Garret, any reference of the ''pretty standard'' derivation ? $\endgroup$
    – OneTwoOne
    Dec 19, 2018 at 23:11
  • $\begingroup$ And it seems to me that reuns'representation is for $\frac{\zeta(s+1)}{s}$, for an unspecified radius of convergence. $\endgroup$
    – OneTwoOne
    Dec 19, 2018 at 23:36
  • 3
    $\begingroup$ Titchmarsh? Edwards' book? Patterson's? Iwaniec'? Iwaniec-Kowalski? Karatsuba-Voronin? Ivic? Even Whittaker-and-Watson? I'd bet a dollar that most of these, if not all, include elementary relations like the above. The only thing that might be not entirely elementary is that the Euler-Mascheroni constant does give the constant in the asymptotics of the harmonic sum...? $\endgroup$ Dec 20, 2018 at 0:02

1 Answer 1

2
$\begingroup$

Let $0<\Re(s)<1$ unless otherwise indicated. Then $$\int_0^\infty (\sum_{1 \le n \le x} 1-x) x^{-s-1}\,dx=-\int_0^1 x^{-s}+\int_1^\infty (\sum_{1 \le n \le x} 1-x) x^{-s-1}\,dx.$$ On the right hand side, the first integral equals $1/(s-1)$. The second integral equals $\zeta(s)/s-1/(s-1)$ as can be seen by proving it first for $\Re(s)>1$ and then continuing it analytically to $\Re(s)>0$. Hence $$\int_0^\infty (\sum_{1 \le n \le x} 1-x) x^{-s-1}\,dx=\frac{\zeta(s)}{s},\qquad 0<\Re(s)<1.\tag{1}$$ Now consider $$S(x):=\sum_{1 \le n \le x} \frac1n-\log x-\gamma,\qquad x>0.$$ Note that $S(x)\ll\log x$ when $x<1$, and $S(x)\ll 1/x$ when $x\geq 1$. Then, the remaining part of the integral is, by integration by parts, $$\int_0^\infty S(x)x^{-s}\,dx=\frac{1}{s-1}\int_0^\infty x^{1-s}\,dS(x).$$ Now we argue for the integral on the right hand side similarly as before: $$\int_0^\infty x^{1-s}\,dS(x)=\int_0^{1-} x^{1-s}\,dS(x)+\int_{1-}^\infty x^{1-s}\,dS(x).$$ On the right hand side, the first integral equals $1/(s-1)$. The second integral equals $\zeta(s)-1/(s-1)$ as can be seen by proving it first for $\Re(s)>1$ and then continuing it analytically to $\Re(s)>0$. Hence $$\int_0^\infty x^{1-s}\,dS(x)=\zeta(s),\qquad 0<\Re(s)<1,$$ and therefore $$\int_0^\infty S(x)x^{-s}\,dx=\frac{\zeta(s)}{s-1},\qquad 0<\Re(s)<1.\tag{2}$$ Subtracting $(2)$ from $(1)$, the claimed identity follows.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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