Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

For every $x > 0$, define $$I(x) = \int_1^\infty \left(\{t\} - \frac{1}{2}\right)\frac{x}{e^{xt}-1}\,dt.$$ where $\{x\} = x - \lfloor x\rfloor$ denotes the fractional part of $x$.

How to justify that $I(x)$ converges to the improper integral $$ \int_1^\infty \left(\{t\}-\frac{1}{2}\right)\frac{dt}{t} $$ when $x \searrow 0$ ?

Of course $\dfrac{x}{e^{tx}-1}$ converges to $\dfrac{1}{t}$ for every $t$, but since $$ \int_1^\infty \left|\{t\} - \frac{1}{2}\right|\frac{dt}{t} = \sum_{n=1}^\infty \int_{-1/2}^{1/2} \frac{|u|\,du}{n+\frac{1}{2}+u} \geq \left(\sum_{n=2}^\infty \frac{1}{n}\right)\int_{-1/2}^{1/2}|u|\,du = +\infty, $$ there is no hope to apply Lebesgue's theorem.

I noticed that this is some type of Abelian-like result. Here is a proof.

Let $f(t)=\left(\{t\}-\dfrac{1}{2}\right)\dfrac{1}{t}$. It is known that $F(T) = \int_1^T f(t)\,dt$ converges to $\ell = \dfrac{1}{2}\ln(2\pi)-1$ as $T$ tends to $+\infty$. Using the following integration by parts formula:

$$\int_T^\infty f(t)\frac{tx}{e^{tx}-1}\,dt = -(F(T)-\ell)\frac{Tx}{e^{Tx}-1}-\int_T^\infty (F(t)-\ell)\frac{d}{dt}\frac{tx}{e^{tx}-1}\,dt$$ where $\dfrac{d}{dt}\dfrac{tx}{e^{tx}-1}=\dfrac{xe^{tx}(1-tx-e^{-tx})}{(e^{tx}-1)} \leq 0$, we get $$ \left|\int_T^\infty f(t)\dfrac{tx}{e^{tx}-1}\,dt\right| \leq 2\sup_{t\geq T}|F(t)-\ell|\dfrac{Tx}{e^{Tx}-1}\leq 2 \sup_{t\geq T}|F(t)-\ell|. $$

Hence, for every $T > 1$ and $x >0$, $$ \left|\int_1^\infty f(t)\dfrac{xt}{e^{xt}-1}\,dt - \ell\right| \leq \left|\int_1^T f(t)\left(1-\dfrac{xt}{e^{xt}-1}\right)\,dt\right| + 3\sup_{t\geq T}|F(t)-\ell|. $$ The result easily follows from this inequality.

share|cite|improve this question
@Clayton It is not obvious that the limit and integral can commute. In fact, I would assume that, that is the what the OP wants to know why. – user17762 Dec 23 '12 at 19:27
@Clayton One cannot simply commute the integral and the limit. – Eric Naslund Dec 23 '12 at 19:37
up vote 5 down vote accepted

Let $$I_0 = \int_{1}^{\infty} \left(\left\{t\right\} - \dfrac12 \right) \dfrac{dt}{t}$$ We then have \begin{align} \vert I(x) - I_0 \vert & = \left \vert \int_{1}^{\infty} \left(\left\{t\right\} - \dfrac12 \right) \left(\dfrac{x}{e^{xt}-1} - \dfrac1t\right) dt\right \vert\\ & = \left \vert \int_{1}^{\infty} \left(\left\{t\right\} - \dfrac12 \right) \left(\dfrac{x}{e^{xt}-1} - \dfrac1t\right) dt\right \vert\\ & = \left \vert \int_{1}^{\infty} \left( \dfrac12 -\left\{t\right\} \right) \left(\dfrac{e^{xt}-1-xt}{t(e^{xt}-1)}\right) dt\right \vert\\ \end{align} Now note that $$\left(\dfrac{e^{xt}-1-xt}{t(e^{xt}-1)}\right) = x\left(\dfrac{e^{xt}-1-xt}{xt(e^{xt}-1)}\right) = x f(xt)$$ where $f(a) = \dfrac{e^a-1-a}{a(e^a-1)}$. Note that $\lim_{a \to 0^+} f(a) =\dfrac12$, $f(a)$ is a decreasing function and $\lim_{a \to +\infty} f(a) = 0$ $$\int_{n}^{n+1/2} \left(\dfrac12 - \{t\}\right)x f(xt) dt = \int_{n}^{n+1/2} \left(\dfrac12 - t + n\right)x f(xt) dt\\ = x f(xt_{n1}) \int_{n}^{n+1/2} \left(\dfrac12 - t + n\right) dt= \dfrac{x f(xt_{n1})}8$$ where $t_{n1} \in [n,n+1/2]$ by MVT. $$\int_{n+1/2}^{n+1} \left(\dfrac12 - \{t\}\right)x f(xt) dt = \int_{n+1/2}^{n+1} \left(\dfrac12 - t + n\right)x f(xt) dt\\ = x f(xt_{n2}) \int_{n+1/2}^{n+1} \left(\dfrac12 - t + n\right) dt = -\dfrac{x f(xt_{n2})}8$$ where $t_{n2} \in [n+1/2,n+1]$ by MVT.

Hence, $$\int_{1}^{\infty} \left( \dfrac12 -\left\{t\right\} \right) \left(\dfrac{e^{xt}-1-xt}{t(e^{xt}-1)}\right) dt = \dfrac{x}8 \left( f(x_{01}) - f(x_{02}) + f(x_{11}) - f(x_{12}) + f(x_{21}) - f(x_{22}) + \cdots \right)$$ Now using the alternating test, since $f(a)$ is a decreasing function converging to $0$,$$\left( f(x_{01}) - f(x_{02}) + f(x_{11}) - f(x_{12}) + f(x_{21}) - f(x_{22}) + \cdots \right)$$ converges to a finite number, say $L$. Hence, we have $$\int_{1}^{\infty} \left( \dfrac12 -\left\{t\right\} \right) \left(\dfrac{e^{xt}-1-xt}{t(e^{xt}-1)}\right) dt = \dfrac{xL}8$$ Now take limit as $x \to 0$ to finish it off.

share|cite|improve this answer
+1, very nice solution. – Eric Naslund Dec 23 '12 at 20:19

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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