Let $\alpha:(0, +\infty) \to (0, +\infty)$ continuous and increasing, with $\alpha(1) = 1$. Show that $\int_1^x d\alpha/\alpha = \log \alpha(x)$ (don't suppose $\alpha$ differentiable)

Well, the change of variable theorem for Stieltjes integration says:

Let $f:[c,d]\to\mathbb{R}$ continuous, $\alpha:[c,d]\to\mathbb{R}$ of bounded variation (rectifiable path) and $\phi:[a,b]\to[c,d]$ an homeomorphism. Then:

$$\int_c^d f(t) d\alpha = \int_a^b f\circ\phi(s)d(\alpha\circ\phi)), \mbox{ if $\phi$ is increasing } $$

$$\int_c^d f(t) d\alpha = -\int_a^b f\circ\phi(s)d(\alpha\circ\phi)), \mbox{ if $\phi$ is decreasing } $$

I think I'm supposed to find the inverse homeomorphism from $\alpha$ (since its continuous and injective), in a way that $\alpha\circ\phi = 1$, so the integral becomes a Riemann integral. I should then evaluate $\int_{\phi(c)}^{\phi(d)} f\circ \phi(s) ds$ where $f: t\to \frac{1}{t}$. The integral now depends on $\phi(s)$, but how to Riemann integrate it?

  • $\begingroup$ I think you want $f(t)=1/\alpha$, $c=1$ and $d=x$. That way the left-hand side in the theorem is your integral $\int_{1}^{x}d\alpha/\alpha$. In that case you are left with integrating $f\circ\phi = 1/t$, which gives the logarithm that you need. $\endgroup$
    – Hellen
    Aug 22 '17 at 21:13
  • $\begingroup$ The catch is that, in your statement of the change of variables theorem, $\phi$ need not be a homeomorphism; it is enough that $\phi$ is continuous and monotone. So you can replace $(f(t), \alpha(t), \phi)$ by $(t^{-1}, t, \alpha)$. $\endgroup$ Aug 23 '17 at 3:47
  • $\begingroup$ But what should be $\alpha\circ\phi$ then? $\endgroup$ Aug 26 '17 at 17:08
  • $\begingroup$ @SangchulLee how can you choose what $\alpha$ is going to be? The exercise gives a specirfic $\alpha$, but you don't know what it is $\endgroup$ Aug 31 '17 at 0:52
  • $\begingroup$ I am not sure what are you asking. The role of $\alpha$ in your problem and the $\alpha$ in the statement are different. That is why I am suggesting to replace $(f(t), \alpha(t), \phi)$ in the theorem by $(t^{-1}, t, \alpha)$, where this $\alpha$ is now what you have in your problem. $\endgroup$ Sep 3 '17 at 21:20

Here is a simple change of variables result:

Claim. Let $f : [c, d] \to \mathbb{R}$ be Riemann integrable and $\alpha : [a, b] \to [c, d]$ be continuous and monotone increasing. Then

$$ \int_{a}^{b} f(\alpha(x)) \, d\alpha(x) = \int_{\alpha(a)}^{\alpha(b)} f(y) \, dy. $$

Proof. For each $\epsilon > 0$, pick $\delta > 0$ such that $|\alpha(x) - \alpha(y)| < \epsilon$ whenever $|x-y| < \delta$. Then for any partition $\Pi = \{a = x_0 < \cdots < x_n = b\}$ with $\|\Pi\| < \delta$ and for any $i = 1, \cdots, n$, we have

$$ m_i [\alpha(x_i) - \alpha(x_{i-1})] \leq \int_{x_{i-1}}^{x_i} f(\alpha(x)) \, d\alpha(x) \leq M_i [\alpha(x_i) - \alpha(x_{i-1})] $$


\begin{align*} m_i &= \inf \{ f(y) : y \in [\alpha(x_{i-1}),\alpha(x_i)] \} \\ M_i &= \sup \{ f(y) : y \in [\alpha(x_{i-1}),\alpha(x_i)] \}. \end{align*}

This tells that, with $\alpha(\Pi) = \{ \alpha(x_0), \cdots, \alpha(x_n)\}$ we have $\|\alpha(\Pi)\| < \epsilon$ and

$$ L(f, \alpha(\Pi)) \leq \int_{a}^{b} f(\alpha(x)) \, d\alpha(x) \leq U(f, \alpha(\Pi)), $$

where $L(f, \cdot)$ and $U(f, \cdot)$ are the lower Riemann sum and the upper Riemann sum, respectively. Therefore the desired identity follows by taking $\epsilon \to 0$.


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.