Recently, I found the following functional equation: $$ n^{nx-1}\cdot\prod_{k=0}^{n-1}{\Gamma{\left(x+\frac{k}{n}\right)}}=\Gamma{(nx)}\cdot\prod_{k=1}^{n-1}{\Gamma{\left(\frac{k}{n}\right)}} $$ Now my question: I've read, that the gamma function is defined as solution of the following three functional equations/conditions: $$ f(x+1)=x\cdot f(x) \space \forall x\in\mathbb R^+ $$ $$ f(1)=1 $$ $$ \ln(f(x)) \space \text {is convex $\forall x\in\mathbb R^+$} $$ And, due to the Bohr-Mollerup theorem, the gamma function is the only function satisfying those three conditions.

Is it equivalent to the following definition?: $$ f(x+1)=x\cdot f(x)\space \forall x\in\mathbb R^+ $$ $$ f(1)=1 $$ $$ n^{nx-1}\cdot\prod_{k=0}^{n-1}{f{\left(x+\frac{k}{n}\right)}}=f{(nx)}\cdot\prod_{k=1}^{n-1}{f{\left(\frac{k}{n}\right)}}\space \forall x\in\mathbb R^+,n\in\mathbb N $$ Meaning, is the gamma function also the only solution to those three functional equations?


These equations do not determine the Gamma function.

Here are some solutions, other than $\Gamma$:

\begin{align*} f_1(x) &= \begin{cases} \Gamma(x) & x\in\mathbb{Q} \\ 0 & x\not\in\mathbb{Q} \end{cases} \\ f_2(x) &= \begin{cases} 2^{\alpha}\cdot\Gamma(x) & \text{if } x=\alpha\sqrt2+\beta; \,\,\alpha,\beta\in\mathbb{Q} \\ \Gamma(x) & x\not\in\mathbb{Q} \end{cases} \\ f_3(x) &= e^{\cos(2\pi x)} \cdot \Gamma(x) \end{align*}

The third example shows that assuming continuity is not sufficient. The assumption about the logarithmic convexity is so important.

  • $\begingroup$ I believe that your last example incorrect - it doesn't satisfy the $n=2$ case of the equation. $\endgroup$ – Alexander Shamov Oct 19 '14 at 6:55
  • $\begingroup$ Yes, the third one is wrong. $\endgroup$ – user141614 Oct 19 '14 at 12:53

As user141614 pointed out correctly, there are other solutions to this equation, so from now on I assume that $f$ is continuous.

The only continuous real-valued $f$ that satisfies the equations is $\Gamma$.

However, it will follow from the proof that by dropping the requirement $f(1) = 1$ we get more interesting solutions, like, e.g., $f(x) = \Gamma(x) (1 - \cos 2 \pi x)^{\alpha}, \alpha > 0$.

Proof. The first thing we show is that $f$ cannot have zeroes.

$f(x) = 0$ for some $x \in \mathbb{N}$, together with the functional equation, would contradict the requirement that $f(1) = 1$.

Assume that $f(x) = 0$ for some $x \notin \mathbb{N}$. Then $f(nx \operatorname{mod} 1) = 0$ for all $n \in \mathbb{N}$. This cannot happen because the orbit $\{nx \operatorname{mod} 1, n \in \mathbb{N}\}$ is either dense or contains an integer.

Now denote $L(x) := \log f(x) - \log \Gamma(x)$. $L$ is a $1$-periodic continuous function that satisfies the equation $$L(nx) = \sum_{k=0}^{n-1} L \left( x + \frac{k}{n} \right) + C_n,$$ where $C_n$ are constants.

Now consider the Fourier series of $L$: $L(x) \sim \sum_{m} L_m e^{2 \pi i m x}$. When rewritten in terms of $L_m$, the equation above becomes $$\sum_m L_m e^{2 \pi inmx} = \sum_m L_m e^{2 \pi imx} \sum_{k=0}^{n-1} e^{2 \pi imk/n} + C_n$$ Note that $\sum_{k=0}^{n-1} e^{2 \pi imk/n}$ is $0$ if $m$ is not divisible by $n$, and $n$ otherwise. So finally we get $L_{nk} = \frac{1}{n} L_k, k \in \mathbb{Z} \setminus 0$. In other words, $$L(x) = L_0 - L_1 \log(1 - e^{2 \pi i x}) - L_{-1} \log (1 - e^{-2 \pi i x})$$

This can only be continuous if $L_1 = L_{-1} = 0$.


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.