I know with integration by parts the answer is $x\ln(x) - x$ but I was wondering how to do this without integration by parts.
-
5What do you mean by 'without integration by parts'? One way would be to differentiate $x\ln(x)-x$ and see that you get $\ln(x)$; you haven't technically used integration by parts, all you've used is the fundamental theorem of calculus and the product rule (but that's all IBP is anyway, just more generally). – Clive Newstead Jul 13 '19 at 12:21
-
3I think there is no other way. The inspiration of the integration by parts is precisely to be able to integrate this type of functions. – azif00 Jul 13 '19 at 12:23
-
Is there any motivation to avoid integration by parts in such cases ? – Peter Jul 13 '19 at 12:40
-
2$$\int\ln(x)dx = \int\big[\ln x + 1 - 1\big]dx = \int(\ln x +1 )dx - \int dx = \int d(x\ln x ) -\int dx = x\ln x - x+c $$ which is almost the same as my previous comment and @Michael's answer. – 19aksh Jul 13 '19 at 12:46
8 Answers
In order to prove $\int \log x \, \mathrm{d}x = x \log x - x + \mathsf{C}$, it suffices to show that
$$ \int_{1}^{x} \log t \, \mathrm{d}t = x \log x - x + 1. $$
We establish this with different methods, avoiding integration by parts technique and 'guessing the antiderivative' strategy.
Method 1. Assume $a \geq 1$. Then by Fubini's theorem,
\begin{align*} \int_{1}^{a} \log x \, \mathrm{d}x &= \int_{1}^{a} \int_{1}^{x} \frac{1}{t} \, \mathrm{d}t\mathrm{d}x \\ &= \int_{1}^{a} \int_{t}^{a} \frac{1}{t} \, \mathrm{d}x\mathrm{d}t \\ &= \int_{1}^{a} \left( \frac{a}{t} - 1 \right) \, \mathrm{d}t \\ &= a \log a - a + 1 \end{align*}
Similar computation shows that the above result continues to hold for $0 < a < 1$.
(In reality, however, this computation still bears the flavor of integration-by-parts technique.)
Method 2. (Regularizing) It can be shown that $(x^{\epsilon} - 1)/\epsilon$ converges uniformly to $\log x$ as $\epsilon \to 0^+$ on any compact subset of $(0, \infty)$. Then
$$ \int_{1}^{x} \log t \, \mathrm{d}t = \lim_{\epsilon \to 0^+} \int_{1}^{x} \frac{t^{\epsilon} - 1}{\epsilon} \, \mathrm{d}t = \lim_{\epsilon \to 0^+} \left( \frac{x^{\epsilon+1}-1}{\epsilon(\epsilon+1)} - \frac{x-1}{\epsilon} \right) = x\log x - x + 1. $$
Method 3. (Solving functional equation) Define the function $f : (0, \infty) \to \mathbb{R}$ by $f(x) = \int_{1}^{x} \log t \, \mathrm{d}t$. Then for $a, b > 0$,
\begin{align*} f(ab) &= \int_{1}^{a} \log t \, \mathrm{d}t + \int_{a}^{ab} \log t \, \mathrm{d}t \\ &= \int_{1}^{a} \log t \, \mathrm{d}t + \int_{1}^{b} a \log (as) \, \mathrm{d}s \tag{$t = as$}\\ &= f(a) + a(b-1) \log a + a f(b). \end{align*}
By switching the role of $a$ and $b$, we also get $f(ab) = f(b) + b(a-1)\log b + bf(a)$, and so,
$$ f(a) + a(b-1)\log a + af(b) = f(b) + b(a-1)\log b + b f(a). $$
Rearranging the identity, for $a, b \neq 1$ we get
$$ \frac{f(a) - a\log a}{a-1} = \frac{f(b) - b\log b}{b-1}. $$
The right-hand side can be further simplified, using $f(1) = 0$ あand $\log 1 = 0$, as
$$ = \frac{f(b) - f(1)}{b-1} - \frac{\log b - \log 1}{b-1} - \log b. $$
Together with $f'(1) = \log 1 = 0$ and $(\log x)'|_{x=1} = 1$, this converges to $-1$ as $b \to 1$. So it follows that
$$ f(a) = a \log a - (a - 1). $$
- 181,930
-
The last method contains in the last step the derivative of $b\log(b)$, which amounts to the same application of product rule/IBP that the OP wanted to avoid. – Lutz Lehmann Jul 13 '19 at 13:44
-
@LutzL, I fixed Method 3 a bit so that it now bypasses the use of product rule by simple algebraic manipulation. – Sangchul Lee Jul 13 '19 at 14:09
-
Could you eliminate the need to differentiate at all if you observe that $h(a)=h(b)$ for all $a,b> 1$ means that $h$ is constant and then insert $f(a)=a\ln(a)+k(a-1)$ back into the first equation for $f$ to determine the constant? – Lutz Lehmann Jul 13 '19 at 14:23
-
@LutzL, That was actually my first attempt, but unfortunately we can check that any $f(x)=x\log x+k(x-1)$ solves the original functional equation. So the equation alone cannot uniquely determines $f$, and we need extra conditions that should follow from the original definition of $f$. – Sangchul Lee Jul 13 '19 at 14:28
-
@Sangchul Lee I think the best way it's just to calculate $(x\ln{x}-x)'$. If you'll see don-votes it's not main. – Michael Rozenberg Jul 13 '19 at 14:43
-
@MichaelRozenberg : If you read the question and initial comments again you will see that OP explicitly asked for alternatives to that approach. It's a bit silly, but still such alternatives exist. – Lutz Lehmann Jul 13 '19 at 14:47
-
@MichaelRozenberg, I have no doubt that differentiating $x\log x - x$ is one of the neatest way of establishing the identity. My answers aim at providing alternative (and less transparent) methods under the assumption that OP may be interested in doing otherwise. – Sangchul Lee Jul 13 '19 at 14:47
-
@Sangchul Lee I meant that $x\ln{x}-x+C$ we can get without an integration by parts, just by the definition of the integral. I really don't understand why we need these methods, which you showed. – Michael Rozenberg Jul 13 '19 at 14:53
-
@MichaelRozenberg, You don't necessarily need these to establish $\int\log x,\mathrm{d}x=x\log x-\log x+\mathsf{C}$. The whole point of this answer is to avoid both integration-by-parts and product-rule and see what else we can do for it. Although product rule is not explicitly mentioned in OP, it has already been pointed out in earlier discussions (see the first comment in OP, for instance) that they are essentially equivalent, suggesting that OP may possibly want to avoid both. This is the motivation of this answer. – Sangchul Lee Jul 13 '19 at 15:04
Compute it via the Riemann sums over the subdivision $x_k=aq^k$ with $x_N=aq^N=b$. Then \begin{align} \int_a^b\ln(x)dx=\lim_{N\to\infty}\sum_{n=0}^{N-1}(\ln(a)+n\ln(q))aq^n(q-1) \end{align} and using the sum formulas for geometric sums $\sum_{n=0}^{N-1}q^n=\frac{q^N-1}{q-1}$ and $\sum_{n=0}^{N-1}nq^n=\frac{q-Nq^N+(N-1)q^{N+1}}{(q-1)^2}$ we get \begin{align} \sum_{n=0}^{N-1}(\ln(a)+n\ln(q))aq^n(q-1) &=a(q^N-1)\ln(a)+a\frac{q-Nq^N+(N-1)q^{N+1}}{q-1}\ln(q) \\ &=(b-a)\ln(a)+\frac{qa-Nb+(N-1)qb}{q-1}\ln(q) \\ &=b(\ln(a)+N\ln(q))-a\ln(a)+\frac{q(a-b)}{q-1}\ln(q) \\ &=b\ln(b)-a\ln(a)-q(b-a)\frac{\ln(q)-\ln(1)}{q-1} \end{align}
Now for $N\to\infty$ we get $q\to 1$ and thus $\ln(q)\to0$, and in the last term the difference quotient converges to the derivative $1$ in $q=1$, so that $$ \int_a^b\ln(x)\,dx=b\ln(b)-a\ln(a)-(b-a) $$ remains.
- 131,652
Graph $y=\ln x$. The area under the curve from $x=1$ to $x=x_0$ is an anti-derivative of $\ln x_0$.
Make the rectangle with base on the $x$-axis from $x=0$ to $x=x_0$ and height equal to $\ln x_0$. The area you want is $x_0 \ln x_0$ minus the area above the curve and inside the rectangle. That is:
$$\int_1^{x_0} \ln x \; dx = x_0\ln x_0 - \int_0^{\ln x_0} e^y \; dy.$$
- 33,728
Recall that $\ln x$ is the inverse of the function $e^x$. We have the following identity for the integrals of inverse functions:
$$\int f^{-1}(x)\,dx=xf^{-1}(x)-F\circ f^{-1}(x)+C,$$
where $F$ is the antiderivative of $f$, not $f^{-1}$. Using this, we obtain the integral of $\ln x$ very easily:
$$\int\ln x\,dx=x\ln x-e^{\ln x}+C=x\ln x-x+C$$
- 5,095
-
This formula to integrate the inverse function is obtained precisely with integration by parts. – azif00 Jul 13 '19 at 12:29
-
@Zacky Using (https://math.stackexchange.com/q/701345) there is a method to show that the identity holds that only uses the product rule for differentiation, not integration by parts – csch2 Jul 13 '19 at 12:30
-
1Not necessarily with IBP "openly", as you can check here: https://en.wikipedia.org/wiki/Integral_of_inverse_functions Yet it is true that this is way more involved than simple IBP...This is what I was writing when the answer popped up. +1 – DonAntonio Jul 13 '19 at 12:30
-
2But the product rule for differentiation is precisely equivalent to IBP. – David C. Ullrich Jul 13 '19 at 12:32
-
@DavidC.Ullrich And that's an excellent way to justify its introduction as a very basic technique. Still, it seems to be the OP wants otherwise. – DonAntonio Jul 13 '19 at 12:33
Hint:
Try writing the Taylor series expansion of $\ln(x)$ at $x=1$ (this involves two cases), integrate it and then compare it with the Taylor series expansion of $x\ln(x)-x$.
PS: But to think of this tricky solution, it is very important to know the answer already (which really doesn't makes sense). Therefore, its always better to use a hammer for a nail instead of using an axe.
- 3,764
Integrate the power series:
\begin{align} \int \log x\, dx &= \int \sum_{i=1}^{\infty} (-1)^{i+1} \frac{(x-1)^i}{i} \\ &= \sum_{i=1}^{\infty} (-1)^{i+1} \int \frac{(x-1)^i}{i} \\ &= \sum_{i=1}^{\infty} (-1)^{i+1} \frac{(x-1)^{i+1}}{i(i+1)} \\ &= (x-1)\sum_{i=1}^{\infty} (-1)^{i+1} \frac{(x-1)^i}{i(i+1)} \\ &= (x-1)\sum_{i=1}^{\infty} (-1)^{i+1} \Big(\frac{(x-1)^i}{i} - \frac{(x-1)^i}{i+1} \Big)\\ &= (x-1) \log x - \sum_{i=2}^{\infty} (-1)^{i} \frac{(x-1)^{i}}{i} \\ &= (x-1)\log x + (\log x - x+1) \\ &= x \log x-x + C. \end{align}
- 483
Just calculate $$\left(x\ln{x}-x\right)'.$$
- 203,855
-
So what, and how did he find this result? You can do that only if you already know the result or you assume what should it be. – nonuser Jul 13 '19 at 13:46
-
1
$$\dfrac{d(x^m\ln x)}{dx}=x^{m-1}+mx^{m-1}\ln x$$
$$\implies x^m\ln x=\int x^{m-1}\ dx+m\int(x^{m-1}\ln x)dx$$
Set $m=1$
- 279,016
-
1To repeat a comment from another answer: But the product rule for differentiation is precisely equivalent to IBP, just putting it under integral signs. So what you wrote is exactly the method of partial integration. – Lutz Lehmann Jul 13 '19 at 13:39