7

$\forall x\in\mathbb R$, $|x|\neq 1$ $$f\left(\frac{x-3}{x+1}\right)+f\left(\frac{3+x}{1-x}\right)=x$$Find $f(x)$.

Now what I'm actually looking for is an explanation of a solution to this problem. I haven't really ever had any experience with such equations.


The solution:

Let $t=\frac{x-3}{x+1}$. Then $$f(t)+f\left(\frac{t-3}{t+1}\right)=\frac{3+t}{1-t}$$

Now let $t=\frac{3+x}{1-x}$. Then $$f\left(\frac{3+t}{1-t}\right)+f(t)=\frac{t-3}{t+1}$$

Add both equalities: $$\frac{8t}{1-t^2}=2f(t)+f\left(\frac{t-3}{t+1}\right)+f\left(\frac{3+t}{1-t}\right)=2f(t)+t$$

Hence the answer is $$f(x)=\frac{4x}{1-x^2}-\frac{x}{2}$$


This is unclear to me. For instance, how come we can assign a different value to the same variable? Does anyone understand this? I'd appreciate any help.

user26486
  • 11,331
  • 2
    The problem should be stated more clearly: The solution suggests that the equation you have given for $f$ holds for all $x ≠ \pm 1$ not just for some $x ≠ \pm 1$. Then the question is to find $f$. (Otherwise it would be impossible, I think.) – k.stm Mar 04 '14 at 19:34
  • 2
    @k.stm I've edited the question. – user26486 Mar 04 '14 at 19:38
  • 2
    If $g(x)=\tfrac{x-3}{x+1}$, $h(x)=\tfrac{3+x}{1-x}$ and $i(x)=x$, then ${i,g,h}$ is a group with composition operator, because $gh=hg=i$, $g^2=h$, and $h^2=g$. Now the assumption is $$fg+fh=i$$ so $$g=ig=fg^2+fhg=fh+fi=fh+f,$$ and in a similar way $$h=f+fg,$$ now add both equalities $$g+h=2f+fg+fh=2f+i,$$ so we have $$f=\frac{g+h-i}{2}.$$ – Woria Mar 04 '14 at 21:02
  • 1
    @Woria, you ought to copy your comment as an answer. – Sammy Black Mar 04 '14 at 21:38
  • @SammyBlack It is too complicated for my elementary understanding, though; e.g. I don't have an idea what a composition operator is. – user26486 Mar 04 '14 at 21:47

5 Answers5

3

I don't know how they came up with this solution; but it is correct.

Look at the first line: For all but finitely many $t$ the number $x:={3+t\over 1-t}$ is admissible, and one has ${3+x\over 1-x}={t-3\over t+1}$ and ${x-3\over x+1}=t$. Since your functional equation is true for all but finitely many $x$ it follows that $$f(t)+f\left(\frac{t-3}{t+1}\right)=\frac{3+t}{1-t}\tag{1}$$ is true for all but finitely many $t$. Similarly, the equation $$f\left(\frac{3+t}{1-t}\right)+f(t)=\frac{t-3}{t+1}\tag{2}$$ is true for all but finitely many $t$.

In $(1)$ and $(2)$ we can write as well $x$ in place of $t$. Adding these equations (with $t$ replaced by $x$) and comparing with the original functional equation we now can say that $$2f(x)+x={8x\over 1-x^2}$$ for all but finitely many $x$. It follows that $$f(x)=\frac{4x}{1-x^2}-\frac{x}{2}\tag{3}$$ for all but finally many $x$.

But it's not over yet: We only have proven that any solution to the original functional equation is of the form $(3)$ for most $x$. We now have to test whether $(3)$ is actually a solution. This is a simple verification which I can leave to you.

  • Well it's quite obvious how, you just substitute for the first and then for the second term inside $f$... – user2345215 Mar 04 '14 at 21:11
  • What do you mean by "for all but finitely many"? – user26486 Mar 04 '14 at 21:14
  • Could you give me a proof that we have to test whether it's a solution? Why do we have to do that? And why "for most $x$"? – user26486 Mar 04 '14 at 21:46
  • @mathh: You have to test the solution because you only prove if $f$ satisfies the equation, then it's of the form bla bla, but that's not a proof it satisfies the equation, because you already assume it. You have to prove the converse too, that's why you need to verify it. – user2345215 Mar 04 '14 at 22:08
  • @mathh: Concerning "all but finitely many": I didn't want to bother with $t$- or $x$-values where a denominator becomes zero. Your second question has already been answered by user2345215. – Christian Blatter Mar 05 '14 at 09:04
1

This is one way to think about it:

For any $t ≠ \pm 1$ you can write $t$ as $t = \tfrac{x-3}{x+1}$ or as $t = \tfrac{3+x}{1-x}$ if you set $x = \tfrac{3+t}{1-t}$ or $x = \tfrac{t-3}{t+1}$ respectively (which means that $z ↦ \tfrac{z-3}{z+1}$ and $z ↦ \tfrac{3+z}{1-z}$ are really involutons inverse functions on $ℝ \setminus \{\pm 1\}$). In your solution, the $x$’s of the two fractions are actually different ones, the $t$ stays the same.

So you can translate your solution to

Let $t ≠ \pm 1$.

Write $t = \tfrac{x_1-3}{x_1+1}$ (such $x_1$ exists), then $f(t) + …$

Write $t = \tfrac{3+x_2}{x_2-1}$ (such $x_2$ exists), then $f(\tfrac{3+t}{1-t}) + …$

The conclusions are still true because the functional equations still hold for the $x_i$’s you used to write $t$ (you have to check they are not $\pm 1$, though). Then you can safely add both equalities without contradiction.


Also, it is worth mentioning that the Mobius transformation $$g \colon ℝ \setminus \{\pm 1\} → ℝ \setminus \{\pm 1\},\, x ↦ \tfrac{x - 3}{x + 1}$$ actually has order $3$, that is $g^3 = \operatorname{id}$, or $g^2 = g^{-1}$, where $g^{-1}$ is actually given by $g^{-1} (x) = \tfrac{3 + x}{1 - x}$. So $g$ and $g^{-1}$ correspond to the fractions you are examining. Then you can think of it that way:

It is given that: $$f∘g + f∘g^{-1} = \operatorname{id}$$ But then, since $g$ has order $3$, you have. \begin{align*} (f + f∘g)∘g &= f∘g + f∘g^2 &=& \operatorname{id}&, \quad \text{and}\\ (f∘g^{-1} + f)∘g^{-1} &= f∘g^{-2} + f∘g^{-1} &=& \operatorname{id}& \end{align*} And so, by multiplying with $g$ or $g^{-1}$ from the right you have: $$f + f∘g = g^{-1}, \quad \text{and} \quad f∘g^{-1} + f = g$$ And so, by adding those and using the functional equation again: $$g + g^{-1} = (f + f∘g) + (f∘g^{-1} + f) = 2f + \operatorname{id},$$ from which you can derive your result $f = \frac{g + g^{-1} - \operatorname{id}}{2}$.

k.stm
  • 18,539
  • What is an involution? Haven't heard of that. – user26486 Mar 04 '14 at 20:01
  • 1
    Here’s wiki on Involution. Something like $x ↦ -x$ or $x ↦ 1/x$ on $x ≠ 0$ – doing it twice is doing nothing, i.e. $-(-x) = x$ and $1/(1/x) = x$ for $x ≠ 0$. The same seems to be true here for the fractions you mentioned. (The only thing you need, though, is that these fractions define bijections $ℝ \setminus {\pm 1} → ℝ \setminus {\pm1}$, well you even only need them to be surjective so you can say “write $t = …$”.) – k.stm Mar 04 '14 at 20:02
  • Oh, actually, I did mess up something, I don’t think there are involutions involved. – k.stm Mar 04 '14 at 20:14
  • mathh: Yeah, I miscalculated something, they’re actually inverse to each other (which I should have known since they are Mobius transformations corresponding to the matrices $\left(\begin{smallmatrix} 1 & -3\ 1 & 1 \end{smallmatrix}\right)$ and $\left(\begin{smallmatrix} 1 & 3\ -1 & 1 \end{smallmatrix}\right)$ which are inverse up to a scalar multiplication). – k.stm Mar 04 '14 at 20:23
  • I'm sorry but it's still quite hard to understand this. If we write $t$ both as $\tfrac{x-3}{x+1}$ and as $\tfrac{3+x}{1-x}$, doesn't that imply that $\tfrac{x-3}{x+1}=\tfrac{3+x}{1-x}$? And that restricts us to only those $x$ that satisfy this equation. – user26486 Mar 04 '14 at 20:46
  • mathh: Yeah, these $x$’s are somehow “locally bound”, I changed them to $x_1$ and $x_2$ which should make it clearer and I also gave another view point on this, eliminating variables. – k.stm Mar 04 '14 at 20:53
  • 1
    What you've written is too complicated for my elementary understanding. But I understand the solution I've given now and there's no need to further explain anything, thank you for your answer. – user26486 Mar 04 '14 at 21:39
  • Normally, I woudn’t ask this, but I put a lot of time into this answer, so I’m curious, downvoter: Why the hate? – k.stm Mar 04 '14 at 22:21
1

It's a bit unclear, but if you look for what $t$'s each of the equalities hold, it starts to make sense.

Note that if $x\ne-1$ and $t_0\ne1$, then $t_0=\frac{x-3}{x+1}\Longleftrightarrow(x+1)t_0=x-3\Longleftrightarrow t_0+3=x(1-t_0)\Longleftrightarrow x=\frac{3+t_0}{1-t_0}$

So the first equality holds for all $t_0\in\mathbb R\setminus\{1\}$.

Similarly if $x\ne1$ and $t_1\ne-1$, then $t_1=\frac{3+x}{1-x}\Longleftrightarrow(1-x)t_1=3+x\Longleftrightarrow t_1-3=x(1+t_1)\Longleftrightarrow x=\frac{t_1-3}{t_1+1}$

So the second equality holds for all $t_1\in\mathbb R\setminus\{-1\}$.

Now for any $x\in\mathbb R$, $|x|\ne1$ we know that $$f(x)+f\left(\frac{x-3}{x+1}\right)=\frac{3+x}{1-x}$$ by the first equation, substituting $x$ for $t_0$ and $$f\left(\frac{3+x}{1-x}\right)+f(x)=\frac{x-3}{x+1}$$ by the second equation, substituting $x$ for $t_1$.

Now add these to get the solution.

user2345215
  • 16,422
  • I'm sorry but it's still quite hard to understand this. If we write $t$ both as $\frac{x−3}{x+1}$ and as $\frac{3+x}{1−x}$, doesn't that imply that $\frac{x−3}{x+1}=\frac{3+x}{1−x}$? And that restricts us to only those $x$ that satisfy this equation. – user26486 Mar 04 '14 at 20:53
  • @mathh: You don't write $t$ as both those expressions, you do it one at the time and forget what $t$ is equal to every time, imagine 3 different variables $t$, $t'$, $t''$ instead. Do you at least understand why the first equality holds for all $t\ne1$ and the second for all $t\ne-1$? – user2345215 Mar 04 '14 at 20:56
  • I understand it now. – user26486 Mar 04 '14 at 21:07
0

t is a symbolic name for the variable, you can rename $x-3\over x+1$ to z and $3+x \over 1-x$ to y and then rename the z and y in the equations to t...

  • The actual question is why we can rename them to $t$. Wouldn't that imply $\frac{x-3}{x+1}=\frac{3+x}{1-x}$? That's where the problem is. – user26486 Mar 04 '14 at 20:20
  • t is not a free variable, it's bounded in the equation and you can rename it as you wish... you say "for all x the equation(1) holds", then if you rename all x to y you can say "for all y equation(1) holds", the validity of equation doesn't depend on t. – Mahdi Dolati Mar 04 '14 at 21:02
  • t is just a place holder, not a variable, when you say let t=$x-1 \over x+1$ you are renaming it in equation you don't add a new constraint to your equations... – Mahdi Dolati Mar 04 '14 at 21:09
0

Since it works for all $x$ it means that $t$ has to take all the values of the domain,since t will be equal to all of those values,both $$\frac{x-3}{x+1},\frac{x+3}{1-x}$$ Will take all those values.so basically $f(t)$ in first will be equal to the $f(t)$ in other,or you can take that for example $$a=\frac{x-3}{x+1},b=\frac{x+3}{1-x}$$

and for example $f(a)=7a$ then $f(b)=7b$ so just changing the letter b into a it will become the same equation

kingW3
  • 13,496
  • I don’t understand your argument at all. For example, how can $t$ be equal to all the values of (which?) domain? – k.stm Mar 04 '14 at 22:27