This question has been completely reformulated and a new property for the function $f_q$ has been added due to a series of helpful comments by fedja.

Consider the integral from quantum field theory due to F.A. Smirnov (see this MSE post): \begin{align} &\int_{-\infty}^{\infty}\prod_{j=1}^3\Gamma\Bigl(\frac 1 3 -\frac {\alpha-\beta_j}{2\pi i}\Bigr) \Gamma\Bigl(\frac 1 3 +\frac {\alpha-\beta_j}{2\pi i}\Bigr)(3\alpha-\sum\beta_m)e^{-\frac{\alpha}2}d\alpha\nonumber\\ &=\frac{(2\pi \Gamma(\frac 2 3))^2}{\Gamma(\frac 4 3)} \prod_{k\neq j}\Gamma\Bigl(\frac 2 3 -\frac {\beta_k-\beta_j}{2\pi i}\Bigr) e^{-\frac12\sum\beta_m}\sum e^{\beta_m},\quad |\text{Im}~\beta_j|<2\pi/3. \end{align} Some analytical and numerical calculations suggest that it has a $q$-analog of the form \begin{align} &\int_{-\infty}^\infty\prod_{j=1}^3\frac{\Gamma_q\left(\frac13+\frac{x-\beta_j}{2\pi i}\right)}{\Gamma_q\left(\frac23+\frac{x-\beta_j}{2\pi i}\right)}\frac{\sum q^{-\frac{\beta_m}{2 \pi i}}-\left(1+q^\frac13+q^{-\frac13}\right) q^{-\frac{x}{2 \pi i}}}{\prod_{m}\sin\left(\frac{\pi}3-\frac{x-\beta_m}{2 i}\right)}e^{-\frac{x}2}dx\\ &=\frac{-2\pi i}{(q;q)_{\infty }^9}\frac{\Gamma_q^2\left(\frac23\right)}{\Gamma_q\left(\frac43\right)}\frac{q^{5/9}}{(1-q)^2}e^{-\frac{1}{2} (\beta_1+\beta_2+\beta_3)} q^{-\frac{\beta_1+\beta_2+\beta_3}{6 \pi i}}\prod_{k\neq j}\Gamma_q\left(\frac23+\frac{\beta_j-\beta_k}{2\pi i}\right)\\ &\times\frac{\theta_q(q^{\frac{\beta_1-\beta_2}{2\pi i}})}{\sin\frac{\beta_1-\beta_2}{2 i}}\frac{\theta_q(q^{\frac{\beta_2-\beta_3}{2\pi i}})}{\sin\frac{\beta_2-\beta_3}{2 i}} \frac{\theta_q(q^{\frac{\beta_3-\beta_1}{2\pi i}})}{\sin\frac{\beta_3-\beta_1}{2 i}} \cdot f_{q_1}(\beta_1,\beta_2,\beta_3).\tag{5} \end{align} where $q_1=e^{-\frac{4 \pi ^2}{\ln(1/q)}}$, $\Gamma_q$ is the $q$-Gamma function and $$ \theta_q(z)=(q;q)_\infty(z;q)_\infty(q/z;q)_\infty. $$ The exact form of the function $f_q$ is unknown but there is evidence that it has a relatively simple closed form expression in terms of finite combination of theta functions.

Obviously $f_q(x,y,z)$ is symmetric in all three arguments $x,y,z$. It is known also that it has the following properties \begin{align} &f_q(x+c,y+c,z+c)=e^cf_q(x,y,z),\tag{1}\\ \\ &f_q\left(\frac{2\pi i}{3},0,z\right) \begin{aligned}[t] & =\left(e^z-1\right) \frac{\theta_q\left(e^{\frac{4 i \pi }{3}-z}\right)}{\theta_q\left(e^{-z}\right)}\\ & =\left(e^z-1\right)e^{-\frac{\pi i}{3}} q_1^{-\frac{1}{9}+\frac{z}{6 \pi i}}\frac{\theta_{q_1}\left(q_1^{\frac13+\frac{z}{2\pi i}}\right)}{\theta_{q_1}\left(q_1^{\frac{z}{2\pi i}}\right)} \end{aligned} \tag{2}\\ \\ &f_q(x,y,z)=e^x+e^y+e^z\\ &+\frac32\left(e^x+e^y+e^z\right)\left(1+e^{x-y}+e^{y-x}+e^{y-z}+e^{z-y}+e^{z-x}+e^{x-z}\right)q\\ &+i\frac{\sqrt{3}}{2}\left(e^{-x}+e^{-y}+e^{-z}\right)\left(e^{2x}+e^{2y}+e^{2z}\right)q+O(q^2) \tag{3} \\ &f_{q}(x,y,z)= e^{-\frac{\pi i}{3}} q_1^{-\frac29} \small{\frac{\left(e^z-e^y\right) q_1^{\frac{-2 x+y+z}{6 \pi i}}+\left(e^x-e^z\right) q_1^{\frac{x-2 y+z}{6 \pi i}}+\left(e^y-e^x\right) q_1^{\frac{x+y-2 z}{6 \pi i}}}{\Big(1-q_1^{\frac{x-y}{2 \pi i}}\Big) \Big(1-q_1^{\frac{z-x}{2 \pi i}}\Big) \Big(1-q_1^{\frac{y-z}{2 \pi i}}\Big)}\left(1+O(q_1^{\frac13})\right)}\tag{4} \\ \end{align}

Eq. $(2)$ is a condition that the difference of LHS and RHS of $(5)$ is an entire function. $(3)$ has been extracted from numerical calculations. $(4)$ has been found from exact $q$-series representation of the integral assuming $\beta_j\in\mathbb{R}$. However numerical calculations show that asymptotics $(4)$ holds if all differences $|\text{Im}~(\beta_j-\beta_k)|\leq \frac{2\pi}{3}$, and one can check that $(4)$ is consistent with $(2)$.

Q: Can anybody make a good guess from these data $1-4$ what this function $f_q(x,y,z)$ might be?

  • 1
    $\begingroup$ Are you assuming that the only dependence on $q$ is in the index of $\theta_q$? $\endgroup$ – fedja Nov 21 '17 at 13:23
  • 1
    $\begingroup$ @fedja yes, as you see I used the notation $\theta_q(x)=\theta(x;q)$ to simplify the formulas. Here $q$ is the base of theta function, not to be confused with the index of the usual definition of Jacobi theta functions. $\endgroup$ – user82588 Nov 21 '17 at 13:27
  • $\begingroup$ Erm... This doesn't quite agree with the equation you posted in the "motivation", which also includes powers of $q$, $1-q$, and even $\theta_q(q)$. What am I missing? $\endgroup$ – fedja Nov 21 '17 at 13:30
  • $\begingroup$ @fedja I really don't see how this is a problem? As you see the base of $f$ in $(5)$ is $q_1$, so if $f$ is a combination of theta functions one can apply imaginary transformation to all theta functions and write them in base $q$, so the RHS of $5$ is a q-series with base $q$. Did this answer your question? $\endgroup$ – user82588 Nov 21 '17 at 13:37
  • 2
    $\begingroup$ Ah, sorry. Missed the symmetry assumption. Then (assuming my expression is $F(x,y)$), use $F(x/3,y/3)F(y/3,z/3)F(x/3,z/3)$. It is still $0$ with two points fixed. (actually you want more symmetries, but it should get clear that one can do that too) $\endgroup$ – fedja Nov 21 '17 at 13:50

The function $$ f_q(x,y,z)=\sum_{cyc}e^z\frac{\theta_q\left(e^{\frac{2 \pi i}{3}+x-z}\right) \theta_q\left(e^{\frac{2 \pi i}{3}+y-z}\right)}{\theta_q\left(e^{x-z}\right) \theta_q\left(e^{y-z}\right)} $$ satisfies all $4$ conditions and also has been confirmed numerically.

| cite | improve this answer | |
  • $\begingroup$ Has it also been proved to satisfy the integral identity (5) or only checked numerically? $\endgroup$ – Qfwfq Nov 24 '17 at 17:30
  • $\begingroup$ @Qfwfq so far it has been proved only that the difference of RHS and LHS of $(5)$ is an entire function (it is assumed that the integral is generalized so that the identity $(5)$ is valid for arbitrary complex $\beta_j$, for example by deforming the contour of integration so that it separates two certain sets of poles of the integrand). $\endgroup$ – user82588 Nov 24 '17 at 17:52
  • $\begingroup$ The difficulty in applying standard methods of proof here is the fact that the integrand contains $3$ factors $\sin\pi(ix+...)$ in the denominator, as opposed to just $2$ factors in the usual $q$-analogs of Barnes' 2nd lemma (to which $(5)$ is close in nature) and 1st lemma (see Gasper and Rahman's book, formula $(4.4.6)$). $\endgroup$ – user82588 Nov 24 '17 at 17:57

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.