1
$\begingroup$

Context

This question deals with the polynomial invariant denoted by $ H_{n} $ in Maksym Fedorchuk and Igor Pak's 2004 paper Rigidity and polynomial invariants of convex polytopes (sections 7.6 and 9). The paper also introduces the associated polynomial invariant, denoted by $ \tilde H_{n} $, which is the polynomial invariant relating the circumradius and sides of a cyclic polygon, which is also known in the literature as the (generalized) Heron r-polynomial.

Product formula

Let $ n $ be an integer such that $ n \geq3 $. Denote $ \left[ n \right] \equiv \{1,2, \dotsc ,n \} $. The polynomial $ H_{n} \left(x_1, \dotsc ,x_n \right) $ is (supposedly) given by the product $$ \prod {\left( x_{n} - \frac{1}{2} \left(1 - \sum_{\begin{aligned} I \subseteq \left[ n-1 \right] \\ \lvert I \rvert \, \text{even} \, \end{aligned}}{\left( \left( -1 \right)^{\frac{\lvert I \rvert}{2}} \prod_{i \in I}{\kappa_i} \right) \prod_{i \in I}{\sqrt{4x_{i} \left( 1-x_{i} \right)}} \prod_{j \in \left[ n-1\right] \setminus I}{\left( 1-2x_{j}\right)}} \right) \right)} $$ where $ \kappa_{n-1} = 1 $ and the product is taken over all possible ($2^{n-2}$) choices of $$ \kappa_{1}, \dotsc , \kappa_{n-2} \in \{-1,1\}. $$

With some work it's possible to show that this product is a polynomial in $ x_1, \dotsc ,x_n $ with integer coefficients.

This explicit formula is my own and does not appear in the literature as far as I know. Here I do not ask for a proof of this formula for $ H_n $. My question is as follows.

Question

Can one expand the product in such a way which will give "nice"/"concrete" functions for the coefficients of its monomial expansion?

"Some work"

In each factor of the product, the part $$ \frac{1}{2} \left(1 - \sum_{\begin{aligned} I \subseteq \left[ n-1 \right] \\ \lvert I \rvert \, \text{even} \, \end{aligned}}{\left( \left( -1 \right)^{\frac{\lvert I \rvert}{2}} \prod_{i \in I}{\kappa_i} \right) \prod_{i \in I}{\sqrt{4x_{i} \left( 1-x_{i} \right)}} \prod_{j \in \left[ n-1\right] \setminus I}{\left( 1-2x_{j}\right)}} \right) $$ can be written as $$ \frac{1}{2} \left(1 - \prod_{j \in \left[ n-1\right]}{\left( 1-2x_{j}\right)} \right) - \\ \frac{1}{2} \left( \sum_{\begin{aligned} \emptyset \neq I \subseteq \left[ n-1 \right] \\ \lvert I \rvert \, \text{even} \; \; \end{aligned}}{\left(\left( -1 \right)^{\frac{\lvert I \rvert}{2}} \prod_{i \in I}{\kappa_i} \right) \prod_{i \in I}{\sqrt{4x_{i} \left( 1-x_{i} \right)}} \prod_{j \in \left[ n-1\right] \setminus I}{\left( 1-2x_{j}\right)}} \right). $$ Also $$ \begin{aligned} \frac{1}{2} \left(1 - \prod_{j \in \left[ n-1\right]}{\left( 1-2x_{j}\right)} \right) &= \sum_{j \in \left[ n-1\right]}{\left(-2\right)^{j-1} \sum_{\begin{aligned} K \subseteq \left[ n-1 \right] \\ \lvert K \rvert = j \; \; \end{aligned}}{\prod_{k \in K}{x_k}}} \\ &= \sum_{j \in \left[ n-1\right]}{\left(-2\right)^{j-1} \operatorname{e}_{j} \left( x_1, \ldots ,x_{n-1} \right)} \end{aligned} $$ where $ \operatorname{e}_{j} $ is the elementary symmetric polynomial. This is a polynomial in $ x_1, \ldots, x_n $ with integer coefficients.

For any $ \ell \in \left[ n-2 \right] $ the factors of the product can be paired off where in each pair the factors only differ in the choice for $ \kappa_{\ell} $, so only the terms of the sums which have $ \sqrt{4x_{\ell} \left( 1-x_{\ell} \right) }$ as a factor change sign. Now each pair is a difference of two squares, and $ \sqrt{4x_{\ell} \left( 1-x_{\ell} \right) }$ can be factored out from inside one square, so to obtain that the appearance of $ x_{\ell} $ in the expansion depends only on powers of $ 4x_{\ell} \left( 1-x_{\ell} \right) $ and $ 1-2x_{\ell} $. The product is a symmetric function in $ x_1, \ldots, x_{n-1} $ hence the last argument also applies for $ \ell = n-1 $. Thus the product is a polynomial in $ x_1, \dotsc, x_n $. Furthermore, the factor of $ \frac{1}{2} $ is absorbed so the coefficients are integers.

$\endgroup$
8
  • $\begingroup$ I know nothing about this, but wouldn't it make sense to look switch from $x_i$ to $y_i=x_i - \frac 12$ and look for the monomial expansion in $y_i$? $\endgroup$ Dec 22, 2021 at 0:46
  • $\begingroup$ @LevBorisov I suppose you meant $ y_i = x_i + \frac{1}{2} $? I'll give it a try, but I worry that going back from $y_i$ to $x_i$ will add (a lot of) complexity to the functions of the coefficients of the monomial expansion. $\endgroup$ Dec 22, 2021 at 10:04
  • $\begingroup$ I do mean the minus, see Alekseyev's calculation. But what is the reason to look at the monomial expansion in $x_i$, if it would be better after a shift? $\endgroup$ Dec 23, 2021 at 11:41
  • $\begingroup$ The question is quite similar to, although harder than, the previous one: math.stackexchange.com/q/3827015 $\endgroup$ Dec 23, 2021 at 18:55
  • $\begingroup$ @LevBorisov Was your initial aim something like Alekseyev's calculation using trig functions and their exponential representations? Because if so than I didn't see that far ahead. For example I don't understand how the substitution $ 4x_{i} \left(1-x_i \right) \rightarrow 4 \left( y-\frac{1}{2} \right)\left( \frac{3}{2} -y \right) $ makes anything simpler, if taken at face value. To answer your question, it's just aesthetics; I'd be happy to know the functions for the coefficients of the monomial expansion of $ H_n \left( x_1+\frac{1}{2}, \ldots, x_n+\frac{1}{2} \right) $, if nothing else. $\endgroup$ Dec 24, 2021 at 12:34

1 Answer 1

2
+25
$\begingroup$

Just to get the ball rolling. Let's introduce $t_j$ such that $1-2x_j=\cos(t_j)$ and $\sqrt{4x_j(1-x_j)}=\sin(t_j)$. Then \begin{split} H_n(x_1,\dots,x_n) &= \prod_{\kappa_1,\dots,\kappa_{n-2}\in\{-1,+1\}}\left( x_n - \frac12\left( 1 - \frac12\prod_{j=1}^{n-1} \big(\cos(t_j) + I\kappa_j\sin(t_j)\big)-\frac12\prod_{j=1}^{n-1} \big(\cos(t_j) - I\kappa_j\sin(t_j)\big)\right)\right) \\ &=\prod_{\kappa_1,\dots,\kappa_{n-2}\in\{-1,+1\}}\left( x_n - \frac12\left( 1 - \cos\big(\sum_{j=1}^{n-1} \kappa_jt_j\big)\right)\right) \\ &=\prod_{\kappa_1,\dots,\kappa_{n-2}\in\{-1,+1\}}\left( x_n - \sin\big(\frac12\sum_{j=1}^{n-1} \kappa_jt_j\big)^2\right) \end{split} By taking logarithm is may be possible to view the corresponding sum as $2^{n-2}$ times the even part of the function in $t_1,\dots,t_{n-2}$.

$\endgroup$
2
  • $\begingroup$ I suppose that the strategy should be to either expand the product as a polynomial in $ x_n $ with coefficients being the elementary symmetric functions in all the different $ \sin $'s; or to look at the sum of monomials with literals either $ x_n $ or a $ \sin $. Then convert the $ \sin $'s to their exponential representations and start looking for patterns, and if such are found, then hopefully use the logarithm to move into polynomials in the $ t_j $'s. When you say "corresponding sum" do you mean the monomial expansion of $ H_n $? And if so, what do you mean when you say "function" later? $\endgroup$ Dec 24, 2021 at 12:55
  • $\begingroup$ Logarithm turns the product into the sum, and this is what I meant under the "corresponding sum". $\endgroup$ Dec 24, 2021 at 15:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.