Kontsevich and Zagier define a period as an integral of a rational function (over $\mathbb{Q}$) defined on a $\mathbb{Q}$-semialgebraic set. They conjecture that if two periods are equal, then the reason may be only a finite number of change of variables and integrations by parts. The question is whether there is a clever computer system which checks whether two periods are equal and if they are, provide a proof (which is the aforementioned sequence.)

A particular motivation is the following identity \begin{align*} 7\int_0^1\log x\cdot \frac{1+x-x^2+x^3-x^4-x^5}{1-x^7}dx=\\ 12\int_{\frac{\sqrt{21}-5}2}^1 \frac{\log|x|}{2+3x+2x^2}dx. \end{align*}

Both parts are periods as may be easily seen if we expand $$\log |x|=\int_{|x|}^1 y^{-1}dy = p.v.\int_{x}^1 y^{-1} dy$$ and so get integrals of rational functions over triangles (if we are ok with p.v.) or triangle/quadrilateral.

Richard Stanley reminds us in this answer at MathOverflow that this is still a conjecture. It comes from the volumes of hyperbolic tetrahedra and it may be proved that the ratio of two sides is a rational number(!), which is calculated and coincides with 1 with accuracy of about 20000 decimal digits.

It is a bit annoying if neither a general theory nor programs motivated by this theory can provide a way to verify such a seemingly toy identity.

  • $\begingroup$ is there a reason that one log has ||, while the other has not? $\endgroup$ – Dima Pasechnik Mar 11 '15 at 11:56
  • $\begingroup$ @DimaPasechnik in LHS x is always positive. Or you mean philosophical reason? $\endgroup$ – Fedor Petrov Mar 11 '15 at 12:08
  • $\begingroup$ oops, sorry. It's $\sqrt{21}-5$. I need better glasses. By the way: integrating polynomials over simplices is just differentiation (of the Laplace transform). One might be tempted to say that same holds for rational functions (which will need to be expanded...). I don't know if this is correct though (convergence is not obvious...). $\endgroup$ – Dima Pasechnik Mar 11 '15 at 12:12
  • $\begingroup$ How about absolute convergence (a requirement in the definition of a period) of your double integrals on respective triangles? Is it obvious? $\endgroup$ – Dima Pasechnik Mar 11 '15 at 13:10
  • $\begingroup$ In the LHS yes, in the RHS principal value becomes absolutely convergent after cancelling the sum $f(x,y)+f(x,-y)$ in small triangle corresponding to negative values of $x$. $\endgroup$ – Fedor Petrov Mar 11 '15 at 13:28

Too long for a comment.

Mathematica produces the LHS equals $$ \frac{1}{7} \left(-\psi ^{(1)}\left(\frac{1}{7}\right)-\psi ^{(1)}\left(\frac{2}{7}\right)+\psi ^{(1)}\left(\frac{3}{7}\right)-\psi ^{(1)}\left(\frac{4}{7}\right)+\psi ^{(1)}\left(\frac{5}{7}\right)+\psi ^{(1)}\left(\frac{6}{7}\right)\right) $$ whereas the RHS is equal to $$ -\frac{12 i \left(\text{Li}_2\left(-\frac{1}{4} i \left(-3 i+\sqrt{7}\right)\right)-\text{Li}_2\left(\frac{1}{4} i \left(3 i+\sqrt{7}\right)\right)+\text{Li}_2\left(\frac{2 i \left(-5+\sqrt{21}\right)}{-3 i+\sqrt{7}}\right)-\text{Li}_2\left(-\frac{2 i \left(-5+\sqrt{21}\right)}{3 i+\sqrt{7}}\right)+\log \left(\frac{1}{2} \left(5+\sqrt{21}\right)\right) \log \left(-7+7 i \sqrt{3}-5 i \sqrt{7}+3 \sqrt{21}\right)-\log \left(\frac{1}{2} \left(5+\sqrt{21}\right)\right) \log \left(-7-7 i \sqrt{3}+5 i \sqrt{7}+3 \sqrt{21}\right)\right)}{\sqrt{7}} .$$


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.