Someone asked on another site about ways to evaluate $\int \ln x \ dx $ without using integration by parts.

My response was the following:

$$ \begin{align} \int\ln x \, dx & =\int\lim_{t\to 0}\frac{x^{t}-1}{t} \, dx \\ &= \lim_{t\to 0}\frac{1}{t}\int (x^{t}-1) \, dx \\ &=\lim_{t\to 0}\frac{1}{t}\left(\frac{x^{t+1}}{t+1}-x\right)+C \\ &=\lim_{t\to 0}\frac{x^{t+1}-x(t+1)}{t(t+1)}+C \\ &=\lim_{t\to 0}\frac{x^{t+1}\ln x-x}{2t+1}+C \\ &= x\ln x-x+C \end{align}$$

But I don't know how to justify moving the limit outside the indefinite integral.

  • $\begingroup$ Because the limit variable is not the variable that you are integrating with respect to. $\endgroup$ – Ataraxia Jun 17 '13 at 16:39
  • $\begingroup$ @ZettaSuro That argument is not very good.. if that was the case then life would be really easy. $\endgroup$ – AD. Jun 17 '13 at 16:42
  • $\begingroup$ @ZettaSuro Well, because there are plenty of counterexamples showing that the argument is wrong. Suppose for simplicity that we worked on a bounded interval, then we would need some argument telling us more about the convergence - like uniform, bounded, monotonicity etc. $\endgroup$ – AD. Jun 17 '13 at 16:49
  • $\begingroup$ @AD. So the principle of linearity does not necessarily apply? $\endgroup$ – Ataraxia Jun 17 '13 at 16:50
  • $\begingroup$ @ZettaSuro No.. $\endgroup$ – AD. Jun 17 '13 at 16:56

First of all, let's fix the interval of integration. We want to show that for any $0<a<b<\infty$, $$\int_a^b \ln x\,dx = \lim_{t\to 0}\int_a^b \frac{x^t-1}{t}\,dx \tag1$$ As nbubis said, it suffices to prove that we have uniform convergence. To this end, write $x=e^u$ and use the Taylor approximation $$|e^{tu}-1-tu|\le C(tu)^2 \tag2$$ for sufficiently small $tu$. The important point is that $|tu|\le |t|\max(|\log a|, |\log b|)$, which allows us to rewrite (2) as $$|e^{tu}-1-tu|\le \widetilde Ct^2 \tag3$$ for sufficiently small $t$, with $\widetilde C$ independent of $u$. Hence, $$ \sup_{x\in [a,b] } \left|\frac{x^{t}-1}{t} -\ln x\right|= \sup_{u\in [\log a,\log b]} \left|\frac{e^{tu}-1}{t} -u\right| \le \widetilde Ct $$ as desired.


I've seen it said that if the series of functions $f_n$ converges uniformly (not only pointwise!) to the function $f$, then you can write:

$$\lim_{n\to \infty} \int f_n dx = \int \lim_{n\to \infty}f_n dx = \int fdx$$ I'll try and find the source, and add it here later.


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.