I'm trying to get the set of solutions of the following equation, whose unknowns are the rational functions $f$ and $g$ :

$\forall x\in\mathbb{R}$ such that the LHS and RHS are both defined, $$f(x)f(x+1)=\frac{g(x+1)}{g(x)}$$

$\left(f(x):=1, g(x):=1\right)$ is an example of a trivial solution to the equation. Now I'm trying to characterize the entire set of solutions.

Since $f$ and $g$ are rational functions, we can write them as $f(x)=\frac{P_1(x)}{Q_1(x)}$ and $g(x)=\frac{P_2(x)}{Q_2(x)}$, with $P_1, P_2, Q_1, Q_2\in\mathbb{R}[X]$.

So this equation is equivalent to finding the set of polynomials $P_1, P_2, Q_1, Q_2$ such that :

$$\forall x,\hspace{6pt}\frac{P_1(x)P_1(x+1)}{Q_1(x)Q_1(x+1)}=\frac{P_2(x+1)Q_2(x)}{P_2(x)Q_2(x+1)}$$

I don't really have any ideas as of how to move forward...

Any suggestion ?


Edit: The answer has been modified following a comment of Harmonic Sun and to include a second, simpler proof.

Any solution is essentially of the form in Ivan Neretin's answer: $$ f(x)=\pm\frac{R(x+1)}{R(x)},g(x)=\pm R(x+1)R(x), $$ where $R(x)$ is an arbitrary rational function. Observe that if a solution $f,g$ with complex coefficients is given we construct $R$ with complex coefficients such that $ f(x)=\pm\frac{R(x+1)}{R(x)},g(x)=R(x+1)R(x).$

Proof: I will first work in the complex setting and then consider the real case. For a rational function $h$ and some complex $\mu$, we define $v(h,\mu)=m$ if $m$ the multiplicity of $x-\mu$ as a factor of $h$. Recall that any rational function can be written as $c\prod_{j=1}^r (x-\mu_j)^{m_j}$. For a rational function $h$ we denote by $h^+$ the one with $h^+(x)=h(x+1)$. Then clearly $v(h^+,\mu-1)=v(h,\mu)$ for all $h$, $\mu$ and $v(hr,\mu)=v(h,\mu)+v(r,\mu)$, $v(h/r,\mu)=v(h,\mu)-v(r,\mu)$ for all $h,r,\mu$.

We introduce another notation for rational functions $h$ and complex $\mu$: $$S(h,\mu)=\sum_{k=-\infty}^\infty v(h,\mu+k).$$ Observe that the sum is essentially finite, that is only a finite number of terms is non-zero. We have $S(h^+,\mu)=S(h,\mu-1)=S(h)$ for all $h$, $\mu$ and $S(hr,\mu)=S(h,\mu)+S(r,\mu)$, $S(h/r,\mu)=S(h,\mu)-S(r,\mu)$ for all $h,r,\mu$.

Now we consider two rational functions $f,g$ such that $$f(x)f(x+1)=\frac{g(x+1)}{g(x)}.$$ Claim 1: $S(f,\mu)=0$ for all $\mu$.

Proof: we calculate on the one hand $S(ff^+,\mu)=S(g^+/g,\mu)=S(g^+,\mu)-S(g,\mu)=0$, on the other hand $S(ff^+,\mu)=S(f,\mu)+S(f^+,\mu)=2S(f,\mu)$.

Claim 2: There exist a rational function $R(x)$ and a constant $c\in\{\pm1\}$ such that $f(x)=c\frac{R(x+1)}{R(x)}$.

$\renewcommand{\Re}{\mbox{Re}\,}$ Proof: We consider the finite set $M$ of complex numbers $\mu$ such that $0\leq \Re\mu<1$ and there exists an integer $m$ such that $v(f,\mu+m)\neq0$. Then $\beta,\tilde\beta$ with $v(f,\beta)\neq0$ and $v(f,\tilde\beta)\neq0$ such that $\tilde\beta-\beta$ is an integer are associated to only one element of $M$. We write $$f(x)=c\prod_{\mu\in M}\prod_{k=-\infty}^\infty (x-\mu-k)^{m_{\mu,k}}$$ with some constant $c$ and integers $m_{\mu,k}$. Again, only a finite number of $m_{\mu,k}$ is nonzero and, interpreting $(x-\alpha)^0=1$ as usual, the product is essentially finite.

Consider now for $\mu\in M$ the integers $n_{\mu,k}=\sum_{\ell=-\infty}^{k-1} m_{\mu,\ell}$ for $k\in\mathbb{Z}$. By Claim 1, we have $\sum_{k=-\infty}^\infty m_{\mu,k}=0$ for all $\mu\in M$ and therefore $n_{\mu,k}$ is non-zero only for a finite number of $k$. Therefore $$R(x)=\prod_{\mu\in M}\prod_{k=-\infty}^\infty (x-\mu-k)^{n_{\mu,k}}$$ defines a rational function. By construction, we have $n_{\mu,k+1}-n_{\mu,k}=m_{\mu,k}$ for all integers $k$. Since $$R(x+1)=\prod_{\mu\in M}\prod_{k=-\infty}^\infty (x+1-\mu-k)^{n_{\mu,k}}= \prod_{\mu\in M}\prod_{k=-\infty}^\infty (x-\mu-k)^{n_{\mu,k+1}},$$ we calculate $$c\,R(x+1)/R(x)=c\prod_{\mu\in M}\prod_{k=-\infty}^\infty (x-\mu-k)^{n_{\mu,k+1}-n_{\mu,k}}= c\prod_{\mu\in M}\prod_{k=-\infty}^\infty (x-\mu-k)^{m_{\mu,k}}=f(x).$$ This proves Claim 2 except for the determination of the constant $c$. Observe here, that $\frac{g(x+1)}{g(x)}$ can be written as a product of powers $(x-\mu)^k$ without a leading constant factor, i.e. leading factor 1. Since $f(x+1)f(x)=\frac{g(x+1)}{g(x)}$, the same is true for $f(x+1)f(x)$. Now this constant factor is also $c^2$. Hence $c^2=1$ and therefore $c\in\{\pm1\}$.

Claim 3: Any $R(x)$ from Claim 2 satisfies $g(x)=\lambda\,R(x+1)R(x)$ with some constant $\lambda$.

Proof: We calculate $f(x)f(x+1)=\frac{R(x+2)}{R(x)}=\frac{h(x+1)}{h(x)}$ where $h(x)=R(x)R(x+1)$. Since also $f(x)f(x+1)=\frac{g(x+1)}{g(x)}$, the quotient $q(x)=\frac{g(x)}{h(x)}$ satisfies $1=\frac{q(x+1)}{q(x)}$ or $q(x+1)=q(x)$. Since $q(x)$ is a rational function, this is only possible if it is a constant, say $q(x)=\lambda$. Hence $g(x)=\lambda R(x)R(x+1)$. In the complex setting, we can now choose a complex $\gamma$ such that $\gamma^2\lambda=1$. Replacing $R(x)$ by $\gamma R(x)$ provides $R$ such that $f(x)=\pm R(x+1)/R(x)$ and $g(x)=R(x+1)R(x)$.

This completes the proof in the complex setting. In the real setting, the essential consideration is to make sure that given real rational $f,g$, the construction of the proof of Claim 2 leads to a real rational $R(x)$. Now if $f(x)$ is real, then $\mu\in M$ if and only if the conjugate $\bar\mu\in M$ and the multiplicities $m_{\mu,k}=m_{\bar\mu,k}$ coincide. The same relation then holds for the multiplicities $n_{\mu,k}$ in the construction of $R(x)$. This implies that $R(x)$ also is a real rational function. Claim 3 also holds in the real setting. Then the only difference to the complex case is that we have to take the sign of $\lambda$ into account and therefore can reach either $g(x)=R(x+1)R(x)$ or $g(x)=-R(x+1)R(x)$.

Second Proof: (Only in the real setting) Consider the rational function $q(x)=g(x)/f(x)$. Using the functional equation of $f,g$, it satisfies $$\frac{q(x+1)}{q(x)}=\frac{g(x+1)f(x)}{f(x+1)g(x)}=f(x)^2,\ \ \ q(x+1)q(x)=\frac{g(x+1)g(x)}{f(x+1)f(x)}=g(x)^2.$$ If we can show that $q(x)$ is essentially a square, that is there exists a rational $R(x)$ such that $q(x)=R(x)^2$ or $q(x)=-R(x)^2$, then we are done.

In order to show that, we use the valuation function $v$ introduced in the first proof. For any complex $\mu$, we calculate using $g(x+1)/f(x+1)=f(x)g(x)$ \begin{equation}\begin{array}{rcl}v(g/f,\mu+1)&=&v({g^+}/{f^+},\mu)=v(fg,\mu)=v(g,\mu)+v(f,\mu)\\&\equiv& v(g,\mu)-v(f,\mu)=v(g/f,\mu)\mod 2.\end{array}\end{equation}

Therefore all the integers $v(q,\mu+m)$, $m\in\mathbb Z$, are congruent modulo 2. As these integers are $0$ for suffiently large $m$, they must all be even. This allows the construction of the wanted $R(x)$. Indeed, if $$q(x)=\alpha\prod_{\mu\in\mathcal M}(x-\mu)^{2m_\mu},$$ where $\alpha$ is a real number and $\mathcal M$ denotes the finite set of $\mu$ such that $v(q,\mu)=2m_\mu\neq 0$, then we can choose $\beta$ such that $\alpha/\beta^2\in\{\pm1\}$ and define $R(x)=\beta\prod_{\mu\in\mathcal M}(x-\mu)^{m_\mu}$. Since $q$ has real coefficients, $\mu\in\mathcal M$ if and only if $\bar\mu\in\mathcal M$ and we have $m_\mu=m_{\bar\mu}$. This implies that $R(x)$ also has real coefficients. It satisfies $q(x)/R(x)^2\in\{\pm1\}$ and our statement follows.

| cite | improve this answer | |
  • $\begingroup$ This is an aweseome answer and some excelent work. It will take me some time to read and fully understand it. Some coments though : $\endgroup$ – Harmonic Sun Apr 11 '19 at 23:15
  • $\begingroup$ (1) You conclude that $R$ has to be a real rational function. But if you take $R$ an always imaginary rational function, $f$ and $g$ are both real and solutions of the equation. (2) No need to have a constant $\lambda$ for the general expression of $g$ : if $f(x) = \frac{R(x+1)}{R(x)}$ and $g(x)=R(x)R(x+1)$ with $R$ an arbitrary rational function, the substitution $R=\sqrt{\lambda}R^*$ leads to $f(x) = \frac{R^*(x+1)}{R^*(x)}$ and $g(x)=\lambda R^*(x)R^*(x+1)$ - this is for $\lambda\geq0$, but for $\lambda<0$ one can take take $i\sqrt{\lambda}$. So any solutions of this form can be reached. $\endgroup$ – Harmonic Sun Apr 11 '19 at 23:15
  • $\begingroup$ @HarmonicSun: You are right. In the complex setting, the solutions can be written $f(x)=\pm R(x+1)/R(x), g(x)=R(x+1)R(x)$ with some rational $R(x)$. In the real setting, I want to reach a real $R$ and then solutions can be written $f(x)=\pm R(x+1)/R(x), g(x)=\pm R(x+1)R(x)$. I will edit my solution. I apologize for the complicated product notation. $\endgroup$ – Helmut Apr 12 '19 at 8:52
  • $\begingroup$ @HarmonicSun:The idea is simple, though. A vector $v$ of $n$ integers, the sum of which is 0 can be written as $v=(u,0)-(0,u)$, where $u$ is some vector of $n-1$ integers. $\endgroup$ – Helmut Apr 12 '19 at 8:56
  • $\begingroup$ @HarmonicSun: I have found a second, simpler proof and also included it in my answer together with a few modifications. I am not sure whether I should remove the first proof... $\endgroup$ – Helmut Apr 13 '19 at 8:45

A fairly broad (sub?)set of solutions is given by the following obvious formula (updated): $$f(x)={R(x+1)\over R(x)},\\[4ex] g(x)={R(x)\cdot R(x+1)} $$ where $R(x)$ is an arbitrary rational function.

All previous iterations of this answer seem to have been covered by this formula.

| cite | improve this answer | |
  • $\begingroup$ Nice answer, I did not see it myself, thanks ! $\endgroup$ – Harmonic Sun Apr 3 '19 at 16:43
  • $\begingroup$ However, this does not include all of the solutions. For instance, for any given $\lambda\in\mathbb{R}^*, (f(x):=1, g(x):=\lambda)$ is a solution that is not included in the set of solutions you gave (exept in the particular case $\lambda=1$) $\endgroup$ – Harmonic Sun Apr 3 '19 at 21:22
  • $\begingroup$ You've just found an extension to my solution which I overlooked: we may multiply my $g(x)$ by $\lambda$ and still be fine. Whether that will cover all solutions remains unclear, though I think it still won't. $\endgroup$ – Ivan Neretin Apr 3 '19 at 21:57
  • $\begingroup$ I was right: it won't. $\endgroup$ – Ivan Neretin Apr 3 '19 at 22:11
  • 2
    $\begingroup$ Wait, I've just generalized it a bit more, so the question is now open again. $\endgroup$ – Ivan Neretin Apr 4 '19 at 13:45

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.