3
$\begingroup$

A real-valued function $f$ defined on the $\mathbb{R}\setminus\{1\}$ by \begin{equation*} f(x) = \frac{3x + 1}{x - 1} = 3 + \frac{4}{x - 1} \end{equation*} is invertible, and \begin{equation*} f^{-1}(x) = \frac{x + 1}{x - 3} = 1 + \frac{4}{x - 3} . \end{equation*} Evaluate constants $A$, $B$, and $C$ such that \begin{equation*} f(2x) = \frac{Af(x) + B}{f(x) + C} \end{equation*} for every real number $x$ distinct from $1/2$ and 1.

This is a problem from a past high school competition given in a county - Monroe County - in New York. (I will post a solution to the problem.) Does every rational function that is equal to the quotient of two linear functions have such a property? Can $f(3x)$ be expressed analogously? How about $f(kx)$ for any positive integer $k$? Is there any advantage to such an expression over substituting "$2x$" in for "$x$"?

  • 1
    Since you added the history tag: What you have here is an *addition formula* expressing $f(x+y)$ as a function of $f(x)$ and $f(y)$ (here for $x=y$). Investigating which functions have (algebraic) addition formulas was a big topic in mathematics starting from early 1800s with Abel and Jacobi (building on earlier works by people like Euler and Fagnano) and getting it's main resolution with Weierstrass whose addition theorem says that if an analytical function have an addition formula it must be a composition of an algebraic function with the identity, the exponential or an elliptic function.2017-02-01
  • 0
    Interesting that you and @Winther see this as a duplication-formula. As you’ll see below, there’s no more mystery in getting the corresponding tripling, $k$-folding, etc. -formulas. You’ll see too that it’s just as easy to represent $f(1/x)$ as an expression in $f(x)$.2017-02-01

4 Answers 4

5

Rational functions of the form $x\mapsto \frac{ax+b}{cx+d}$ (with $ad\ne bc$) form a group under composition (Möbius transformations). You are given such an $f$ (and its inverse $f^{-1}$) and want some $g$ such that $g\circ f=f\circ 2$ (where the duplication function $2(x)=\frac{2x+0}{0x+1}$ is a Möbius transformation as well). We apply $\circ f^{-1}$ from the right to find $$ g=f\circ 2\circ f^{-1}.$$ In this particular case, $$g(x)=f\left(2\cdot \frac{x+1}{x-3}\right) =\frac{3\cdot2\cdot \frac{x+1}{x-3}+1 }{2\cdot \frac{x+1}{x-3}-1}=\ldots =\frac{7x+3}{x+5}$$

As long as we stay within Möbius transformations , there is nothing special about the $2$ in this problem. It would not be more complicated to ask how to express $f(\frac{x+42}{2-x})$ in terms of $f(x)$.

  • 0
    I appreciate you couching an explanation in the context of group theory.2017-02-01
1

This question is easily answered, IF you know the relation between fractional-linear transformations and $2$-by-$2$ matrices.

A fractional-linear transformation is, as you’ve identified, quotient of two linear polynomials that doesn’t reduce to a constant, so representable as $x\mapsto\frac{ax+b}{cx+d}$. The wonderful fact, verifiable with a minimum of work, is that these transformations are well representable by matrices, $\begin{pmatrix}a&b\\c&d\end{pmatrix}=M$ in this case. If $\,f$ and $g$ are frac. lin. tfs., represented by matrices $M$ and $N$ as above, then the wonderfulness is that $f\circ g$ is represented by $MN$. You must be aware, however, that if $\lambda$ is a nonzero constant, $\lambda M$ and $M$ represent the same fractional-linear transformation.

You may not recognize $x\mapsto2x$ as a fractional-linear transformation, but it is, and its matrix representation is $P=\begin{pmatrix}2&0\\0&1\end{pmatrix}$. Your original frac. lin. tf. is $f$, with matrix $Q=\begin{pmatrix}3&1\\1&-1\end{pmatrix}$. Looking at the problem finally, you see that you are being asked to find $R$ such that $RQ=QP$. Solve for $R$ in noncommutative matrix-land, and you find that $R=QPQ^{-1}$. Fortunately, they’ve given you $Q^{-1}$. Thus $$ R=\begin{pmatrix}3&1\\1&-1\end{pmatrix} \begin{pmatrix}2&0\\0&1\end{pmatrix} \begin{pmatrix}1&1\\1&-3\end{pmatrix}= \begin{pmatrix}7&3\\1&5\end{pmatrix}\,, $$ from which you read off your coefficients.

To a student armed with the theory, it should take about two minutes to finish this question off.

0

Here is the intended solution to the problem. This is a lot of work for a problem on a ten-minute competition. (There are six topics at each competition, and in each topic, there are three problems. The problem in the post comes from the topic called "Functions." I doubt anyone solved this problem during the competition.)

Solution

If $x > 1$, for example, there is exactly one real number $y > 3$ such that $f^{-1}(y) = x$. \begin{equation*} f(2x) = \frac{Ay + B}{y + C} . \end{equation*} $2x > 1$, $f(2x) > 3$, and $f(2x)$ is in the domain of $f^{-1}$. \begin{equation*} \frac{2y + 2}{y - 3} = 2x = f^{-1}\bigl(f(2x)\bigr) = f^{-1}\left(\frac{Ay + B}{y + C}\right) = \frac{\dfrac{Ay + B}{y + C} + 1}{\dfrac{Ay + B}{y + C} - 3} = \frac{(A + 1)y + B + C}{(A - 3)y + B - 3C} . \end{equation*} \begin{equation*} (2y + 2)\bigl[ (A - 3)y + B - 3C \bigr] = (2A - 6)y^{2} + (2A + 2B - 6C - 6)y + 2B - 6C , \end{equation*} and \begin{equation*} (y - 3)((A + 1)y + B + C) = (A + 1)y^{2} + [-3A + B + C - 3]y - 3B - 3C . \end{equation*} For every real number $y > 3$, \begin{equation*} (2A - 6)y^{2} + (2A + 2B - 6C - 6)y + 2B - 6C = (A + 1)y^{2} + (-3A + B + C - 3)y - 3B - 3C . \end{equation*} So, $(A, \, B, \, C)$ is the solution to the system of linear equations \begin{equation*} \begin{cases} 2x - 6 = x + 1 \\ 2x + 2y - 6z - 6 = -3x + y + z - 3 \\ 2y - 6z = -3y - 3z \end{cases} \end{equation*} in the variables $x$, $y$, and $z$. The only solution to this system of equations is $(7, \, 3, \, 5)$. So, $A = 7$, $B = 3$, and $C = 5$.

0

Let $\displaystyle f(x)=\frac{Ax+B}{Cx+D}$, $x\in\Bbb{R}\setminus\{-\frac DC\}$. For every $k\in\Bbb{N}$, we assume there exists $P,Q,T\in\Bbb{R}$ satisfy

$$f(kx)=\frac{Pf(x)+Q}{f(x)+T}$$

Then $$f(kx)[f(x)+T]=Pf(x)+Q$$

$$\frac{Akx+B}{Ckx+D}\cdot\frac{(A+CT)x+(B+DT)}{Cx+D}=\frac{(AP+CQ)x+(BP+DQ)}{Cx+D}$$ $$(Akx+B)\left[(A+CT)x+(B+DT)\right]=(Ckx+D)[(AP+CQ)x+(BP+DQ)]$$

Comparing both sides we get \begin{equation*} \begin{cases} A(A+CT)=C(AP+CQ) \\ B(A+CT)+Ak(B+DT)=D(AP+CQ)+Ck(BP+DQ) \\ B(B+DT)=D(BP+DQ) \end{cases} \end{equation*}

We have to solve the system of $P,Q,T$:

$$ \left[ \begin{array}{ccc|c} AC&C^2&-AC&A^2\\ (AD+kBC)&CD(k+1)&-(BC+kAD)&AB(k+1)\\ BD&D^2&-BD&B^2\\ \end{array} \right] $$

Finally we get $$Q=-\frac{AB}{CD},T=\frac{(\frac AC+\frac BDk)\beta-2AB(k+1)}{(k-1)\alpha},P=T+\frac{\beta}{CD}$$

where $\alpha=AD-BC$ and $\beta=AD+BC$.

If the solution exists, the conditions are $$CD\neq0,\alpha\neq0,k\neq1$$

  • 0
    Thanks for doing the computations for me. I was not in the mood to do them again.2017-02-01
  • 0
    I am really curious what prompted a mathematician to express $f(2x)$ in terms of $f(x)$. I added the tag "math history" to my post.2017-02-01
  • 0
    Since the computation is massive, I would appreciate if someone can check the result and comment on my mistakes.2017-02-01