Why are the moments of my random variable infested with Fibonacci polynomials?

69 Views Asked by At

Preamble

I've stumbled upon (a particular variant of) Fibonacci polynomials in my work, defined by $$ F_k(x) = \begin{cases} \qquad\qquad 1, & k=1 \\ \qquad\qquad 1, & k=2 \\ F_{k-1}(x) + x\!\; F_{k-2}(x), & k > 2 \end{cases} $$ giving rise to a sequence of polynomials starting with $$\begin{aligned} \begin{split} F_1(x) &= 1 \\ F_2(x) &= 1 \\ F_3(x) &= 1 + x \\ F_4(x) &= 1 + 2x\\ \end{split} &\qquad\qquad \begin{split} F_5(x) &= 1 + 3x + x^2\\ F_6(x) &= 1 + 4x + 3x^2 \\ F_7(x) &= 1 + 5x + 6x^2 + x^3 \\ F_8(x) &= 1 + 6x + 10x^2 + 4x^3 \end{split} \end{aligned}$$ and so forth. I'm very pleased to have encountered them, but I'd like to know what business they think they have in the moments of a random variable ranging over $[0,1]$ that I'm looking at.

Context

I'm doing work with a random variable $X$ which ranges within $[0,1]$, with moments $$ \mathbb E [X^d] \,=\, 1\big/(d+1) \,.$$ Using $X$, I define another variable $Y_t$ depending on a real parameter $t \in [0,1]$: $$ Y_t \;=\; {(1-t) X + t(1-X)} \;=\; {t + (1-2t)X} $$ Computing the first few moments of $Y_t$ explicitly, what I found was that they form integer polynomials in $(t^2 - t)$, divided by $(d+1)$. After looking up the sequence of coefficients of the integer polynomials at the OEIS, I encountered A011973, which allowed me to realise that the polynomials governing these moments could be written as $$ \mathbb E[Y_t^d] = \tfrac{1}{d+1} F_{d+1}(t^2 - t), \qquad 0 \leqslant d \leqslant 4. $$

Question.

This is all very nice, but I only have circumstantial evidence: I don't have any proof that the moments of $Y_t$ have anything to do with Fibonacci polynomials. But the formula $F_{d+1}(t^2 - t)/(d+1)$ recovers the exact lower and upper bounds which I know holds of the moments of $Y_t$, and structures like this rarely come up by accident in my line of work.

What business does a recursively generated polynomial sequence like this have doing in the moments of my random variable?

1

There are 1 best solutions below

0
On

$$\begin{align*} \mathbb E[Y_t^d] &= \mathbb E\left[\sum_{k=0}^d \binom{d}{k} (1-2t)^k X^k t^{d-k} \right] \\ &= \sum_{k=0}^d \binom{d}{k} (1-2t)^k \mathbb E[X^k] t^{d-k} \\ &= \sum_{k=0}^d \binom{d}{k} (1-2t)^k t^{d-k} \cdot \frac{1}{k+1} \\ &= \sum_{k=0}^d \frac{d!}{(k+1)!(d-k)!} (1-2t)^k t^{d-k} \\ &= \sum_{k=0}^d \frac{1}{d+1} \cdot \frac{(d+1)!}{(k+1)!((d+1)-(k+1))!} (1-2t)^k t^{d-k} \\ &= \frac{1}{(d+1)(1-2t)} \sum_{k=0}^d \binom{d+1}{k+1} (1-2t)^{k+1} t^{(d+1)-(k+1)} \\ &= \frac{1}{(d+1)(1-2t)} \sum_{k=1}^{d+1} \binom{d+1}{k} (1-2t)^k t^{(d+1)-k} \\ &= \frac{1}{(d+1)(1-2t)} \left( -\binom{d+1}{0} (1-2t)^0 t^{(d+1)-0} + ((1-2t) + t)^{d+1} \right) \\ &= \frac{(1-t)^{d+1} - t^{d+1}}{(d+1)(1-2t)}. \end{align*}$$ To show equivalence to your choice of definition of Fibonacci polynomials, you need only establish that the recurrence holds; i.e. let $$G_d(x) = \frac{(1-x)^d - x^d}{1-2x}.$$ We aim to show $$F_d(t^2-t) = G_d(t).$$ If this is the case, then $G$ should obey the recurrence $$G_d(t) = G_{d-1}(t) + (t^2-t)G_{d-2}(t).$$ I leave this as a straightforward algebraic exercise, as well as establishing the base cases. This completes the proof.