I understand how the antiderivative of $\log(x)$ can be obtained by Integration by Parts (i.e. product rule), but I was wondering how-if at all- it could be obtained only using sum/difference rule and substitution/chain rule.

  • 1
    $\begingroup$ To the best of my knowledge, avoiding integration by parts all together is not possible. Of course, one can convert your integral into some other integral through a u-sub, but then that integral is to be evaluated without any integration by parts. Doubtful, and I am really looking forward to any answer to make that work. $\endgroup$ – imranfat Sep 29 '16 at 2:16
  • 6
    $\begingroup$ What on earth is happening in the edit history of this question? $\endgroup$ – JiK Sep 29 '16 at 13:24
  • 3
    $\begingroup$ @JiK looks like a conflict between "don't contradict the OP" editors and "OP obviously added irrelevant crap / is trying to ask a second question, try to rescue it" editors. $\endgroup$ – hobbs Sep 29 '16 at 17:00

$$\int_1^t \ln(x)\,dx = \int_1^t\int_1^x\frac 1u\,du\,dx = \int_1^t\int_u^t\frac 1u\,dx\,du = \int_1^t\frac{t-u}{u}\,du = {\large[}t\ln(u)-u{\large]}_{1}^t$$

  • 6
    $\begingroup$ I realize adding an extra dimension and using Fubini's theorem isn't exactly "only using sum/difference rule and substitution/chain rule", but I feel it's still in the spirit of the question. $\ddot \smile$ $\endgroup$ – user137731 Sep 29 '16 at 2:38
  • 1
    $\begingroup$ +1 this is beautifully simple. $\endgroup$ – Simply Beautiful Art Sep 29 '16 at 13:24
  • 1
    $\begingroup$ Silly question - why do the indices of the integration change in that particular way when the integration order is changed? $\endgroup$ – templatetypedef Sep 29 '16 at 16:37
  • 2
    $\begingroup$ @templatetypedef - graph the region of integration in (x,u)-space. You will see that both regions are the same triangle with corners $(1,1), (t,1), (t,t)$. $\endgroup$ – Paul Sinclair Sep 29 '16 at 17:55
  • 1
    $\begingroup$ @templatetypedef: Paul Sinclair gives a good geometric answer; stated more algebraically, it’s because the conditions $1 \leq x \leq t$, $1\leq u \leq x$ are equivalent to the conditions $1 \leq u \leq t$, $u \leq x \leq t$. $\endgroup$ – Peter LeFanu Lumsdaine Sep 29 '16 at 17:59

There's a fun formula relating the integral of an invertible function and the integral of its inverse, namely $$ bf(b)-af(a)=\int_a^bf(t)dt+\int_{f(a)}^{f(b)}f^{-1}(t)dt $$ Which is easily seen from a picture. Then setting $b=x$, $0<x<a$ and $f(t)=\log t$ gives $$ x\log x-a\log a=\int_a^x\log tdt+\int_{\log a}^{\log x}e^tdt\\ \Rightarrow x\log x-a\log a-\int_{\log a}^{\log x}e^tdt=\int_a^x\log tdt\\ \Rightarrow x\log x-x+a-a\log a=\int_a^x\log tdt\\ \Rightarrow x\log x-x+c=\int_a^x\log tdt $$

  • $\begingroup$ Almost right, except I believe the LHS of your first equation should be $bf(b) - af(a)$. After all, we have $\int_1^x \log t\,dt = x \log x - x + 1$. $\endgroup$ – arkeet Sep 29 '16 at 3:28
  • $\begingroup$ @arkeet there, that should also be more general $\endgroup$ – qbert Sep 29 '16 at 3:39
  • $\begingroup$ Sure, although I think it would be a satisfactory answer to just have an answer for one $a$. $\endgroup$ – arkeet Sep 29 '16 at 3:46
  • 2
    $\begingroup$ I need to remember this... $\endgroup$ – Simply Beautiful Art Sep 29 '16 at 13:28

We have:

$$\int_0^{t} x^a dx = \frac{t^{a+1}}{a+1}$$

Differentiating w.r.t. $a$ gives:

$$\int_0^{t} x^a \log(x)dx = \frac{t^{a+1}\left[(a+1)\log(t)-1\right]}{(a+1)^2}$$

The result then follows if we put $a = 0$.


Fermat's Method of Exhaustion can be used to obliterate this problem from first principles.

However, we will require the evaluation of a limit, which will be done so in advance.


$\displaystyle \lim_{r \to 1} \frac{r \log{r}}{1-r} = -1$


Consider the elementary inequality which holds true for all $r>0$:

$\displaystyle r-1 \le \log{r} \le 1-\frac{1}{r}$

The result follows by the squeeze theorem.

Consider a real number $0<r<1$, which will be used to construct the infinite geometric sequence of intervals.

Construct the lines $x = b, x = br, x = br^2, x = br^3, \cdots$ all the way to infinity.

The region bounded by $\log x, x = br^n, x = br^{n+1}$ is bounded strictly from below and above by the following inequality:

$\displaystyle b(1-r)r^{n+1} (\log{b}+(n+1)\log{r}) < \int_{br^{n+1}}^{br^n} \log{x}\, \text{d}x < b(1-r)r^n (\log{b}+n\log{r})$

Summing up these inequalities from $n=0$ to $\infty$ will give us strict upper and lower bounds on the integral.

After some elementary Arithmetico-Geometric Summation, we have the following:

$\displaystyle r \, b \, \log{b} + b\,\frac{r \log{r}}{1-r} < \int_0^b \log{x} \, \text{d}x < b\log{b} + b\,\frac{r \log{r}}{1-r}$

Take the limit as $r$ approaches $1$, and by the squeeze theorem, it can then be concluded:

$\displaystyle \int_0^b \log{x} \, \text{d}x = b\log{b} - b$

  • 1
    $\begingroup$ "Fermat's Method of Geometric Riemann Summation" - a curious name, given that Fermat died in 1665 while Riemann was born in 1826. $\endgroup$ – Paul Sinclair Sep 29 '16 at 18:00
  • $\begingroup$ Nice profile picture (+1) $\endgroup$ – clathratus Sep 12 at 23:33

This may not necessarily be what you're looking for, but I got a kick out of it so I figured I'd share. We may evaluate the antiderivative by using integration by parts indirectly.

We seek the integral $$I_f(x)=\int f^{-1}(x)dx.$$ We set $x=f(u)$ to get $$I_f(x)=\int uf'(u)du.$$ Integration by parts gives $$I_f(x)=uf(u)-\int f(u)du$$ which is $$I_f(x)=xf^{-1}(x)-F\circ f^{-1}(x)+C.$$ Choosing $f^{-1}(x)=\ln x,$ $$I_{\ln}(x)=x\ln x-x+C.$$


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.