7

For $a>0$, prove $$\left(\frac{1}{a}+\frac{1}{2}\cdot\frac{x}{a+2}+\frac{1\cdot 3}{2\cdot 4}\cdot \frac{x^2}{a+4}+\cdots\right) \cdot \left( 1+\frac{1}{2}\cdot x+\frac{1\cdot 3}{2\cdot 4}\cdot x^2 +\cdots \right) = \frac{1}{a} \left[ 1+\frac{a+1}{a+2}x+\frac{(a+1)(a+3)}{(a+2)(a+4)} x^2 +\cdots \right]$$

I'm a bit lost on how to prove it.

First I tried to prove it by induction but the expansion of $x^n$'s coefficient on the left side is complex.

Then I tried to simplify it. Call it $S_1 S_2 = S_3$, apparently $S_2 = \sum_{n\ge 0} 4^{-n}{2n \choose n} x^n = \frac{1}{\sqrt{1-x}}$, but I got problem in $S_1$.

What I have done is:

Denote $c_n = 4^{-n} {2n \choose n}$, then $$\left( \sum_{n\ge 1} \frac{c_n}{n} x^n \right)' = \sum_{n\ge 1} c_n x^{n-1} = \frac1x \left( \sum_{n\ge 0} c_n x^n - 1 \right) = \frac1x\left(\frac{1}{\sqrt{1-x}}-1\right) $$

Let $f(x) = -2 \ln (1+\sqrt{1-x})$, then $f' = \frac1x\left(\frac{1}{\sqrt{1-x}}-1\right)$. Integral from $0$ to $x$ leads to $$\sum_{n\ge 1} \frac{c_n}{n} x^n = \int_0^x \frac1x\left(\frac{1}{\sqrt{1-x}}-1\right) dx = f(x) |_0^x =\ln 4 - 2 \ln (1+\sqrt{1-x})$$

To work out $S_1=\sum_{n\ge 1} \frac{c_n}{2n+a} x^n = \frac12 \sum_{n\ge 1} \frac{c_n}{n+b} x^n$, where $b = a/2$, define $g(x) = 2S_1 = \sum_{n\ge 1} \frac{c_n}{n+b} x^n $. Then we have

$$ \frac{(x^b g)'}{x^b} = g' + \frac{b}{x}g = \frac1x\left(\frac{1}{\sqrt{1-x}}-1\right) $$

So

$$g = \frac{1}{x^b} \left( \int_0^x \frac{x^{b-1}}{\sqrt{1-x}} dx - \frac{x^b}{b} + c \right) $$ , where $c$ is a constant.

But then I'm stuck at the ​$ \int_0^x \frac{x^{b-1}}{\sqrt{1-x}} dx$ integral.

athos
  • 5,177
  • Surely $S_2=\sum_{n\ge 0} 4^{-n} \binom{2n}{n} x^n $, correct? – Vishu Apr 07 '21 at 12:03
  • @Tavish sorry my typo, updated the question. – athos Apr 07 '21 at 12:05
  • 1
    I guess you already tried this: \begin{align} a_n&=\frac{1\cdot 3\cdot\ldots\cdot (2n-1)}{2\cdot4\cdot\ldots\cdot (2n)}\frac{1}{a+2n}\ b_n&=\frac{1\cdot 3\cdot\ldots\cdot (2n-1)}{2\cdot4\cdot\ldots\cdot (2n)} \end{align} Show (at least a try is worthwhile) $$c_n=\sum^n_{k=0}a_nb_{n-k}=\frac{(a+1)\cdot\ldots\cdot(a+2n-1)}{(a+2)\cdot\ldots\cdot(a+2n)}$$ – Mittens Apr 07 '21 at 20:42
  • @OliverDiaz the first two do have cancellations. Seems I need work on generalise it. Thanks. – athos Apr 07 '21 at 20:53
  • 2
    Your formula for $S_2$ is wrong. $S_2=\frac{1}{\sqrt{1-x}}$ – Moko19 Apr 07 '21 at 21:13
  • @Moko19 thanks, i've updated the post. – athos Apr 08 '21 at 17:08

1 Answers1

4

After forming the Cauchy product for the LHS, we see that it suffices to prove that $$\sum_{k=0}^n\frac{c_k c_{n-k}}{a+2k}=\frac1a\prod_{k=1}^n\frac{a+2k-1}{a+2k}\tag{*}\label{coeffs}$$ where $c_n=\displaystyle\prod_{k=1}^n\frac{2k-1}{2k}$ as in the OP.

But \eqref{coeffs} is just the partial fraction decomposition of its RHS: $$f(a):=\frac1a\prod_{k=1}^n\frac{a+2k-1}{a+2k}=\sum_{k=0}^n\frac{b_k}{a+2k},$$ where $b_k$ are constants (that don't depend on $a$, and) determined by $$b_j=\lim_{a\to-2j}(a+2j)f(a)=\frac{\prod_{1\leqslant k\leqslant n}(2k-2j-1)}{\prod_{0\leqslant k\leqslant n,\ k\neq j}(2k-2j)}=c_j c_{n-j}$$ after writing the numerator as $\prod_{k=1}^j\cdot\prod_{k=j+1}^n$ and the denominator as $\prod_{k=0}^{j-1}\cdot\prod_{k=j+1}^n$.

Mittens
  • 39,145
metamorphy
  • 39,111