4

In all textbooks and online notes, there is always a table of antiderivatives and it always says $\int \frac {1}{x} \mathrm dx = \ln(|x|)+C_1$ but there is nowhere a proof. I found some proofs online but there is too much circular logic, assuming unproven hypothesis to reach the conclusion. Differentiating $\ln(x)$ can't be the most rigorous proof out there...

Ali Caglayan
  • 5,726
UserX
  • 4,930
  • What do you think the definition of $\ln x$ is? There are several ways of defining it that all eventually turn out to be equivalent, but the answer to your question depends heavily on which one you choose... – Micah Aug 20 '14 at 19:44
  • 1
    There are two main approaches. You can define $\ln(x) = \int_1^x \frac{1}{y} dy$, in which case the antiderivative follows by a fairly straightforward argument to deal with signs. The other way is to define $\exp(x)$ in any way, prove that $\frac{d}{dx} \exp(x) = \exp(x)$, then define $\ln$ as the inverse of $\exp$ and apply the inverse function theorem. – Ian Aug 20 '14 at 19:45
  • $\ln(x)=\int_1^x \frac {1}{t} dt$, I assume you're talking about this definition, but this is the same as assuming what I want to prove. – UserX Aug 20 '14 at 19:46
  • The last way, of course, is to find the limit directly, which you would do by writing $\frac{\ln(x+h)-\ln(x)}{h}=\frac{\ln \left ( \frac{x+h}{x} \right )}{h} = \frac{\ln \left ( 1+ \frac{h}{x} \right )}{h}$ and then showing (by some other property) that this converges to $1/x$. This is probably the hardest approach. – Ian Aug 20 '14 at 19:57
  • 1
    Aside: just in case you're unaware, $c_1$ is not a constant, but is instead "locally constant": specifically, it has the form $$c_1 = \begin{cases} c_2 & x < 0 \ c_3 & x > 0 \end{cases} $$ where $c_2$ and $c_3$ truly are constants (with respect to $x$). –  Aug 20 '14 at 20:10
  • Perhaps my answer here is of use: http://math.stackexchange.com/questions/852209/proof-that-int-frac1x-is-lnx/2163842#2163842 or here: http://math.stackexchange.com/questions/2118082/why-isnt-int-frac1xdx-fracx00/2118089#2118089 – Simply Beautiful Art Feb 27 '17 at 15:15

6 Answers6

2

$$(\ln(|x|)+C)'=\ln'(|x|)=\frac{|x|'}{|x|}=\frac{\text{sgn}(x)}{|x|}=\begin{cases}\frac{1}{|x|}&if\ x>0\\\frac{-1}{|x|}&if\ x<0\end{cases}=\frac{1}{x}$$

Q.E.D.

idm
  • 11,824
  • 1
    -1 the original poster explicitly states that they do not just want to differentiate the log function. This is not much more than that. – user157227 Aug 20 '14 at 20:29
  • @user157227: Thank you for your comment to justify the -1, I appreciate :-) – idm Aug 20 '14 at 20:46
2

Let $f(x)=\ln x$

Explicitly: $$f'(x)=\lim_{h\to0}\frac{f(x+h)-f(x)}h=\lim_{h\to0}\frac{\ln(x+h)-\ln(x)}h=\lim_{h\to0}\frac{\ln(1+h/x)}{h.(h/x)}.(h/x)=\frac1x$$


EDIT: $$\frac{d}{dx}\ln|x|=\frac1{|x|}.\frac{d}{dx}|x|=\frac1x$$

RE60K
  • 17,716
  • 1
    In fact, you made the proof for $x>0$ only. – idm Aug 20 '14 at 20:10
  • @idm Though, having proven $\frac{\mathrm{d}}{\mathrm{d}x}\ln(x)$, you need only use the chain rule on $\ln\vert,x,\vert$ with $u=\vert,x,\vert$. [Edit: I now realise that your answer covers this though Aditya's answer was halfway there :) ] – Jam Aug 20 '14 at 20:12
2

Hint: For $h\neq0$, we have $~\displaystyle\int_1^xt^{~h-1}~dt=\bigg[\frac{t^h}h\bigg]_1^x=\frac{x^h-1}h.~$ For $h=0$, we have $~\displaystyle\int_1^x\frac{dt}t=$

$=\displaystyle\lim_{h\to0}\frac{x^h-1}h.~$ At the same time, $~\Big(a^x\Big)'=\displaystyle\lim_{h\to0}\dfrac{a^{x+h}-a^x}h=a^x~\lim_{h\to0}\dfrac{a^h-1}h.~$ Do you notice

anything “suspicious” ? ;-) Let e be the number for which this limit is $1$, and let $\ln=\log_{~\large e}.$ Then

it follows that $~\Big(a^x\Big)'=\Big(e^{x\ln a}\Big)'=(x\ln a)'\cdot e^{x\ln a}\cdot1=a^x\ln a.~$ Thus, $~\displaystyle\int_1^x\frac{dt}t=\ln x,$ which

for $x=e,~$ yields $~\displaystyle\int_1^{\large e}\frac{dt}t=1.~$ Now, let us inspect the numbers $u(h)$ for which $~\displaystyle\int_1^{u(h)}t^{h-1}~dt$

$=1.~$ Integrating, we have $\dfrac{u^h-1}h=1\iff u=\sqrt[^h]{1+h}.~$ Letting $h\to0,~$ we have $e=u(0)$

$=\displaystyle\lim_{h\to0}(1+h)^{^\tfrac1h}=\lim_{n\to\infty}\bigg(1+\dfrac1n\bigg)^n$. Hope this helps.

Lucian
  • 48,334
  • 2
  • 83
  • 154
2

We will start by defining

$$\log(x) \equiv \int_1^x \frac{dt}{t}$$

and then showing that this is indeed the inverse function of the exponential function $e^{x}$. By definition and for $x>0$, $\log(x)$ is continious and monotonely increasing with $\log(1) = 0$.

We first start by showing that $\log(x) + \log(y) = \log(xy)$:

$$\log(x) + \log(y) \equiv \int_1^x \frac{dt}{t} + \int_1^y \frac{dt}{t} = \int_1^x\frac{dt}{t} + \int_x^{xy} \frac{d(t/x)}{(t/x)}= \int_1^{xy} \frac{dt}{t} \equiv \log(xy)$$

Where we switched the integration variables $t\to t/x$ in the last integral and used $\int_a^b + \int_b^c = \int_a^c$. By the same method we can show that $\log(x^n) = n\log(x)$:

$$\log(x^n) \equiv \int_1^{x^n} \frac{dt}{t} = \int_1^{x}\frac{ndt^{1/n}}{t^{1/n}} = n\int_1^x \frac{dt}{t} = n\log(x)$$

Since $\log(x)$ is monotonely increasing it has an inverse function $e(x)$ which satisfy

$$\log(e(x)) + \log(e(y)) = \log(e(x)e(y))\to e(x + y) = e(x)e(y)$$

Plugging in $x=y=0$ we have $e(0) = 1$. Further we find

$$\log(e(x)^{\frac{1}{x}}) = \frac{1}{x} \log(e(x)) \to e(x)^{\frac{1}{x}} = e(1)$$

so $e(x) = e(1)^x$. Finally, to show that $e(x) = e^x$ we must calculate $e(1)$ and show that this is really equal to $e$ ($ = 2.7128\ldots$). The definition of $e$ I'm going to use is that $e$ is the unique constant such that

$$\lim_{h\to 0} \frac{e^h-1}{h} = 1$$

since this is the definition that is used to show that $\frac{de^x}{dx} = e^x$. Plugging in $e(1)$ in the definition we get that our constant must satisfy

$$1 = \int_1^{e(1)}\frac{dt}{t}$$

and that $e(1) = e$ follows from Lucians answer above.

Winther
  • 24,478
1

Cop-out answer

Define $\ln(x):=\int_1^x \frac{1}{t}dt$. Then the result follows immediately.

Serious answer

$$ \int \color{green}{\frac{1}{x}}dx = \int \color{green}{\sum_{k=0}^\infty (-1)^k (x-1)^k} dx \\$$ $$= \sum_{k=0}^\infty (-1)^k \int (x-1)^k dx \tag{pulling out the constant }\\ = \sum_{k=0}^\infty \frac{(-1)^k}{k+1} (x-1)^{k+1} + C = \ln |x| +C $$

beep-boop
  • 11,595
  • Unless I'm missing something, this argument only works for $|x-1| < 1$, i.e. $0 < x < 2$. Both series diverge outside that radius. –  Aug 20 '14 at 20:56
  • 1
    Why do you assume the series definition of the log function as given? Finding the series requires the derivative of the log function and much more machinery than is needed in general. – user157227 Aug 20 '14 at 20:58
  • 1
    @alexqwx That's not a safe proof, as a matter of fact, you've proved the identity in the most unsafest way! – FreeMind Aug 20 '14 at 23:33
  • @user157227 You must have something as given for the definition of $\ln x$. – Jean-Claude Arbaut Mar 25 '16 at 08:29
0

Here is the proof I wrote with Ian's guiding. Ian, thank you for pointing me in the right direction for what I was looking for; $ln(x)$ is defined to be the inverse function of $e^x$. The inverse function theorem states that $(f^{-1})'(b)=\frac{1}{(f'(a))}$, $b=f(a)$. Let $f(x)=\ln(x), f^{-1}(f^{-1}(x))=f(x)$, thus, $f^{-1}(x)=e^x$. $(f^{-1})'=(e^x)'=e^x$. Ergo, on the LHS, $(f^{-1})'(f(x))=e^{\ln(x)}=x$. RHS then becomes $\frac{1}{(\ln(x))'}$, thus $x=\frac{1}{(\ln(x))'} \iff x^{-1}=(\frac{1}{(\ln(x))'})^{-1} \iff \frac{1}{x}=(\ln(x))' \iff \int \frac{1}{x} dx= \int( \ln(x))' dx$. Using the Fundamental theorem of calculus, $\int \frac{1}{x} dx=\ln(x)+c$.Then suspect c is $iπ$, and because of the identity $\ln(x)=\ln(-x)+iπ$, $\ln(x)$ becomes $\ln(|x|)$, but I don't know how to prove that.

UserX
  • 4,930
  • The constant of integration is just signifying that integral equations are determined up to a constant. In this case, you do not determine your "c". – user157227 Aug 20 '14 at 20:59
  • Can't think of any other way the $\ln(x)$ becomes $\ln(|x|)$ other than doing some "obvious" assumptions(or looking at the graph), but I hate assumptions in proofs. – UserX Aug 20 '14 at 21:04
  • @user148432: your entire argument, starting with "let $f(x) = \ln(x)$", assumes that $x > 0$. In that domain, $|x| = x$. You will need to modify the argument if it is to work for $x < 0$. –  Aug 20 '14 at 21:07
  • If I add $f : \mathbb{R} \rightarrow \mathbb{C^{*}}$ would it solve the problem? – UserX Aug 20 '14 at 21:14
  • You've shown that (ln(x))' = 1/x, when x > 0. You would need to show that (ln(-x))' = 1/x, when x < 0. Together, this implies that ln(|x|)' = 1/x when x =/= 0. – Alex Zorn Aug 20 '14 at 21:16
  • That sounds like a lot of more work. I'll continue the proof tommorow. Thanks. – UserX Aug 20 '14 at 21:22