18
$\begingroup$

I've been asked to elaborate on the following evaluation:

$$ \begin{align}\\ \displaystyle {\large\int_0^{1}} \!\cfrac 1 {1 + \cfrac 1 {1 + \cfrac 1 {\ddots + \cfrac 1 { 1 + \psi (\left\{1/x\right\}+1)}}}} \:\mathrm{d}x & = \dfrac{F_{n-1}}{F_{n}} - \dfrac{(-1)^{n}}{F_{n}^2} \ln \!\left(\!\dfrac{F_{n+2}-F_{n}\gamma}{F_{n+1}-F_{n}\gamma} \right)\\\\ \end{align} $$ where $\left\{x\right\}=x-\lfloor x\rfloor$ denotes the fractional part of $x$, $\gamma$ is the Euler-Mascheroni constant, $F_{n}$ are the Fibonacci numbers, $\psi:=\Gamma'/\Gamma$ is the digamma function and where the continued fraction has a total of $n$ horizontal bars.

$\endgroup$
26
$\begingroup$

Here is a general approach.

Recall that the digamma function $\displaystyle \psi : = \Gamma'/\Gamma$ admits the following expansion, coming from the Weierstrass infinite product representation of the $\Gamma$ function, $$\begin{equation} \psi(x+1) = -\gamma + \sum_{k=1}^{\infty} \left( \frac{1}{k} - \frac{1}{x+k} \right), \quad x >-1, \end{equation} $$ where $\gamma$ is the Euler-Mascheroni constant. By differentiation, one obtains $$\begin{equation} \psi'(x+1) = \sum_{k=1}^{\infty} \frac{1}{(x+k)^2}, \quad x>-1. \end{equation} $$

Theorem 1. Let $f$ be integrable on $(0,1)$. Then $$\begin{equation} \displaystyle \int_{0}^{1} f \left(\left\{1/x\right\}\right) \mathrm{d}x = \int_{0}^{1} f(x) \: \psi'(x+1) \mathrm{d}x \tag1 \end{equation}$$ where $\left\{x\right\}=x-\lfloor x\rfloor$ denotes the fractional part of $x$, $\psi'$ being the derivative of the digamma function $\displaystyle \psi : = \Gamma'/\Gamma$.

Proof. One may write \begin{align*} \displaystyle \int_{0}^{1} f \left(\left\{1/x\right\}\right) \mathrm{d}x &= \sum_{k=1}^{\infty}\int_{\frac{1}{k+1}}^{\frac{1}{k}} f \left(\left\{1/x\right\}\right) \mathrm{d}x \\ &= \sum_{k=1}^{\infty} \int_{k}^{k+1} f\left(\left\{ u \right\}\right) \: \frac{\mathrm{d} u}{u^{2}} \\ &= \sum_{k=1}^{\infty} \int_{k}^{k+1} f\left( u-k \right) \: \frac{\mathrm{d} u}{u^{2}} \\ &= \sum_{k=1}^{\infty} \int_{0}^{1} f\left( v \right) \: \frac{\mathrm{d} v}{(v+k)^{2}} \\ &= \int_{0}^{1} f\left( v \right) \sum_{k=1}^{\infty} \frac{1}{(v+k)^{2}} \: \mathrm{d} v \\ &= \int_{0}^{1} f(v) \: \psi'(v+1) \: \mathrm{d}v, \end{align*} where the interchange between the infinite sum and the integration is allowed by the uniform bound: $$ \quad \left|\, \sum_{k=1}^{N} \frac{1}{(v+k)^{2}} \, \right| \, < \, \frac{\pi^{2}}{6}, \quad N \geq 1, \, 0 \leq v \leq 1.$$

One of the consequences of Theorem 1 is that the transformation $ x \rightarrow \gamma+\psi ( \left\{1/x\right\}+1)$ leaves the Lebesgue measure on (0,1) invariant.

Theorem 2. Let $f$ be integrable on $(0,1)$. Then $$\begin{align} \displaystyle \int_{0}^{1} f \left(\gamma+\psi ( \left\{1/x\right\}+1)\right)\:\mathrm{d}x = \int_{0}^{1} f(x) \: \mathrm{d}x \tag2 \end{align} $$ where $\left\{x\right\}=x-\lfloor x\rfloor$ denotes the fractional part of $x$, $\gamma$ being the Euler-Mascheroni constant and $\psi:= \Gamma'/\Gamma.$

Proof. From $(1)$, one gets $$ \begin{align*} \displaystyle \int_{0}^{1} f \left(\gamma+\psi (\left\{1/x\right\}+1)\right) \:\mathrm{d}x &= \int_{0}^{1} f \left(\gamma+\psi (x+1)\right) \psi' (x+1) \: \mathrm{d}x \\ & = \int_{0}^{1} f \left(\gamma+\psi (x+1) \right) (\gamma+\psi (x+1))' \: \mathrm{d}x \\ & = \int_{0}^{1} f(u) \: \mathrm{d}u, \end{align*} $$ using the change of variables $u=\gamma+\psi (x+1) $ which gives $u(0)=\gamma+\psi(1)=0$ and $u(1)=\gamma+\psi(2)=1$.

Theorem 2 enables one to evaluate a great variety of integrals involving the digamma function in the integrand.

Proposition 1. Let $n=0,1,2,\cdots .$ Then $$\begin{align} \displaystyle \int_{0}^{1} \left(\psi( \left\{1/x\right\}+1)\right)^{n}\:\mathrm{d}x & = \sum_{k=0}^{n}\frac{(-1)^{k}}{n-k+1}{{n}\choose k}\gamma^{k} \end{align} \tag3 $$ and $$\begin{align} \gamma^{n} = (-1)^{n}\sum_{k=0}^{n}\!{{n}\choose k}B_{k}\!\displaystyle \int_{0}^{1} \! \left(\psi( \left\{1/x\right\}+1)\right)^{n-k}\mathrm{d}x. \end{align} \tag4 $$ where $\gamma$ is the Euler constant, $\psi:= \Gamma'/\Gamma$ and $B_{k}$ are the Bernoulli numbers.

Proof. Using $(1)$, we may prove $(3)$ by writing $$\begin{align*} \int_{0}^{1} \left(\psi( \left\{1/x\right\}+1)\right)^{n} \mathrm{d}x &= \int_{0}^{1} \left(\gamma+\psi( \left\{1/x\right\}+1)-\gamma\right)^{n}\:\mathrm{d}x \\ &=\sum_{k=0}^{n} (-1)^{n-k}\gamma^{n-k}{{n}\choose k} \int_{0}^{1} \left(\gamma+\psi( \left\{1/x\right\}+1)\right)^{k}\:\mathrm{d}x \\ &= \sum_{k=0}^{n} (-1)^{n-k}\gamma^{n-k}{{n}\choose k} \int_{0}^{1} x^{k}\:\mathrm{d}x \\ & = \sum_{k=0}^{n}\frac{(-1)^{k}}{n-k+1}{{n}\choose k}\gamma^{k}.\end{align*}$$ Identity $(4)$ is then deduced from $(3)$ by appealing to an inversion combinatorial sum (J. Riordan, Inverse Relations and Combinatorial Identities. The American Mathematical Monthly, Vol. 71, No. 5, May 1964, p. 495).

Proposition 2. $$ \begin{align}\\ \displaystyle \int_0^{1} \cfrac 1 {1 + \cfrac 1 {1 + \cfrac 1 {\ddots + \cfrac 1 { 1 + \psi (\left\{1/x\right\}+1)}}}} \mathrm{d}x & = \dfrac{F_{n-1}}{F_{n}} - \dfrac{(-1)^{n}}{F_{n}^2} \ln \left( \dfrac{F_{n+2}-F_{n}\gamma}{F_{n+1}-F_{n}\gamma} \right) \tag5 \\\\ \end{align} $$ where $\left\{x\right\}=x-\lfloor x\rfloor$ denotes the fractional part of $x$, $\gamma$ is the Euler constant, $F_{n}$ are the Fibonacci numbers, $\psi:=\Gamma'/\Gamma$ is the digamma function and where the continued fraction has $n$ horizontal bars.

Proof. Using $(2)$, we just have to evaluate the corresponding elementary rational function. One may prove by a simple induction that \begin{align} \displaystyle \cfrac 1 {1 + \cfrac 1 {1 + \cfrac 1 {\ddots + \cfrac 1 { 1 + \, x \,}}}} = \dfrac{F_{n}x+F_{n+1}}{F_{n+1}x+F_{n}} \end{align} where the continued fraction has $n$ horizontal bars, leading easily to $(5)$.

$\endgroup$
  • $\begingroup$ Wow. Long. This probably is one of the longest answers. $\endgroup$ – Joao Dec 10 '14 at 0:42
  • $\begingroup$ Very impressive. $\endgroup$ – marty cohen Jan 9 '15 at 21:21

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.