6
$\begingroup$

Consider this recurrence relation:

$$ \begin{eqnarray*} f_0&=&1\\ f_n&=& \sum_{m=0}^{n-1} \frac{\left(\frac{m+3}{2}\right)_{m-1}}{\left(\frac{m+2}{2}\right)_m} f_{n-m-1} f_m\ \ \ \text{for $1\leq n$.} \end{eqnarray*} $$ where the Pochhammer symbol denotes the rising factorial. The generating function $f(z)=\sum_{n=0}^\infty f_nz^n$ seems to be a root of $$ 0=12 f^3 z^2- (f-1)^2 (f+2) $$ I have checked this to be true for the first 600 terms. However, I have been unable to come up with a proof. Do you have any ideas on how I might show this to be true?

Cheers, Petter

$\endgroup$

3 Answers 3

9
$\begingroup$

This is sequence A244038 in OEIS after scaling by $3^n$, so $f_n=(4/3)^n\binom{3n/2}n$. The fact that it satisfies a cubic equation is certainly a well-known result in hypergeometric functions.

EDIT: remove the "well-known": set $F(x)=\sum_{n\ge0}\binom{3n/2}{n}x^n$. Then $$F(x)={}_2F_1(1/3,2/3;1/2;27x^2/4)+(3x/2)\,{}_2F_1(5/6,7/6;3/2,27x^2/4)$$ (which can probably be slightly simplified using contiguity relations), but can a hypergeometric expert explain why $(27x^2/4-1)F^3+3F-2=0$ ?

$\endgroup$
1
  • $\begingroup$ Thanks a lot! I don't know how the denominators all being powers of three escaped me. Both the above formulas showed up in some QFT calculations so being able to show they are the same is very nice :). $\endgroup$
    – Petter
    Commented Jan 28, 2018 at 0:04
0
$\begingroup$

To prove that $F(z)$ is a root of $(27x^2/4-1)F^3+3F-2=0$, we rewrite as:

[1] $x^3+12x/(27z^2-4)-8/(27z^2-4)=0$

Then making use of equation (48) of the roots of the general trinomial https://arxiv.org/pdf/2212.09919.pdf

Letting $I=0$ one can trivially see that it yields the desired binomial expansion like $F(z)$, but it differs slightly because the denominator has $(n+1)$.

After some labor using (48) we have

$2u^{-1}/3-1/3 \sum_{n\ge0}\binom{3n/2}{n}\frac{u^n}{n+1}$

In which $u=(27/(4-27z^2))^{-1/2}$

Which is a root of [1], which we will denote as $G(z)$

$F(z)$ and $G(z)$ are obviously related by differentiating or integrating.

Converting $F(z)$ and $G(z)$ to their trig equivalents (using binomial to -->> hypergeometric --> inverse-trig transformations):

$\sum_{n\ge0}\binom{3n/2}{n}\frac{u^n}{n+1}= (2- \frac{4}{\sqrt{3}}\cos(\frac{1}{3} \arcsin (\frac{3u\sqrt{3}}{2})+\pi/6) )/u $

$\sum_{n\ge0}\binom{3n/2}{n} z^n = ( 4 \sin(\frac{1}{3} \arcsin (\frac{3z\sqrt{3}}{2})+\pi/6) )\frac{1}{\sqrt{4-27z^2}}$

Thus, we have to prove the following is true:

$\sin(\frac{1}{3} \arcsin (\frac{3z\sqrt{3}}{2})+\pi/6)=\cos(\frac{1}{3} \arcsin (\frac{3u\sqrt{3}}{2})+\pi/6)$

Because $G(z)$ converges for $0<z<2z/3\sqrt{2/3} $ we have:

$ \arcsin (\frac{3z\sqrt{3}}{2})+ \arcsin (\frac{3u\sqrt{3}}{2})=\pi/2 $

Using

$\arcsin x + \arcsin y = \arcsin( x\sqrt{1-y^2} + y\sqrt{1-x^2})$ it is easy to confirm it is true.

$\endgroup$
0
$\begingroup$

Here's a sketch of a proof, using Lagrange inversion, of the equation $(27x^2/4-1)F^3+3F-2=0$, where $$F(x) = \sum_{n=0}^\infty \binom{3n/2}{n}x^n.$$

One form of Lagrange inversion says that if $h(x) = x R(h(x))$ then (writing $h$ for $h(x)$) for any power series $\psi$, $$[x^n] \frac{\psi(h)}{1-xR'(h)} = [t^n]\psi(t) R(t)^n.$$ Since $\binom{3n/2}{n}= [x^n]\bigl( (1+x)^{3/2}\bigr)^n$, taking $R(t) = (1+t)^{3/2}$ and $\psi(t) = 1$ gives $$F =\frac{1}{1-\frac32 x (1+h)^{1/2}}\tag{1}$$ where $h=x(1+h)^{3/2}$ and thus $$h^2 = x^2(1+h)^3\tag{2}.$$ Solving $(1)$ for $h$ in terms of $F$ and $x$, substituting in $(2)$, and factoring gives $$[(27x^2-4)F^3 +12F -9 ][(27x^2-20)F^3+48F^2 -36F]=0\tag{3}.$$ We can check that the second factor on the left side of $(3)$ is not 0 (the coefficient of $x^3$ is $-54$) so the first factor must be 0

It may be noted that if we set $A(x) = F(4x) = \sum_{n=0}^\infty 4^n \binom{3n/2}{n}x^n$, so that the coefficients of $A(x)$ are integers, then in the OEIS page A244038 for the coefficients of $A(x)$, Michael Somos states that $A(x)$ satisfies $A(x)^3 (1 - 108 x^2) = 3A(x) - 2$, which is equivalent to $(27x^2/4-1)F^3+3F-2=0$. Somos also gives a reference to my paper A short proof of the Deutsch-Sagan congruence for connected non crossing graphs, which has some further information about these and related integers.

It remains to prove that proposer's recurrence for the numbers $f_n$ is satisfied by $f_n = (4/3)^n\binom{3n/2}{n}$. This does not seem easy.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .