1: \begin{proof}

2:   By replacing all $a_3,a_4,\ldots,a_h$ to $a_2$ in Equation~(\ref{FilterEquation}), we have

3:   \begin{align*}

4:     y&\geq\sum_{i=1}^h\binom{h}{i}a_2(\alpha x)^i(1-\alpha x)^{h-i}-\binom{h}{1}(a_2-a_1)\alpha x(1-\alpha x)^{h-1}\\

5:     &=a_2\left(\sum_{i=0}^h\binom{h}{i}(\alpha x)^i(1-\alpha x)^{h-i}-(1-\alpha x)^h\right)-h(a_2-a_1)\alpha x(1-\alpha x)^{h-1}\\

6:     &=a_2-a_2(1-\alpha x)^h-h(a_2-a_1)\alpha x(1-\alpha x)^{h-1}\\

7:     &=a_2-a_2\exp(h\ln(1-\alpha x))-h(a_2-a_1)\alpha x\exp((h-1)\ln(1-\alpha x))\\

8:     &\geq a_2-a_2\exp(-h\alpha x)-h(a_2-a_1)\alpha x\exp(-\alpha x(h-1))\tag{concavity of $\ln$ function}.

9:   \end{align*}

10:   Letting $x=a_1\gamma$, we have

11:   \begin{align*}

12:     y-x&\geq a_2-a_2\exp(-\gamma(1-\delta))-(a_2-a_1)(1-\delta)\gamma\exp\left(\gamma(1-\delta)\left(\frac1h-1\right)\right)-a_1\gamma\tag{since $x=a_1\gamma$ and $h\alpha a_1=1-\delta$}\\

13:     &>a_2(1-e^{-\gamma}-\gamma e^{-\gamma})-a_1(\gamma-\gamma e^{-\gamma})-\epsilon,

14:   \end{align*}

15:   where in the last step, for any $\epsilon>0$, we can find small enough $\delta$ and large enough $h$ to make the inequality holds.

16:   Rigorously, we have $1-\delta\rightarrow1$ and $\frac1h\rightarrow0$ for $\delta\rightarrow0$ and $h\rightarrow\infty$. The expression in the second last step is a continuous function, which has limit $a_2(1-e^{-\gamma}-\gamma e^{-\gamma})-a_1(\gamma-\gamma e^{-\gamma})$, and the last step is obtained by the definition of limit.

17: 

18:   Therefore, $y-x>0$ for any $x>p_1$ with $x=a_1\gamma$, where $\gamma$ satisfies

19:   $$a_2(1-e^{-\gamma}-\gamma e^{-\gamma})-a_1(\gamma-\gamma e^{-\gamma})>0,$$

20:   which implies the proposition.

21: \end{proof}