this is for holding javascript data
Wen Jenny Shi edited section_subsectionPr.tex
over 9 years ago
Commit id: ece32fa6ce2cc4d384bcee5a1cf6f817422eae6c
deletions | additions
diff --git a/section_subsectionPr.tex b/section_subsectionPr.tex
index ea23566..1f7b6c6 100644
--- a/section_subsectionPr.tex
+++ b/section_subsectionPr.tex
...
\section{} \section{Supplementary materials}
\subsection{Proof of binary splitting}
In this subsection we consider the simplest case. Suppose there are only two genome positions to be clustered, $Y=[Y_1,\; Y_2]$. If they share the same probability parameter, then the likelihood of the two share the same parameter is one when the numbers of observations at the two sites $m_1$ and $m_2$ are large.
...
Stirling's formula provides the following approximation:
$$\log\Gamma(z)\approx\frac{1}{2}\log(2\pi)-\frac{1}{2}\log $\log\Gamma(z)\approx\frac{1}{2}\log(2\pi)-\frac{1}{2}\log z+z\log
z-z,$$ z-z,$
Therefore,
\begin{eqnarray*}
&&LR\\
...
Under null hypothesis that $Y_1$ and $Y_2$ follow the same distribution, i.e. they share the same probability parameter. Denote the comment probability parameter as $P=(p^1,\cdots,p^5)$. Then the normal approximation of the multinomial random variables are
$$y_i^j\approx $y_i^j\approx m_ip^j+\sqrt{m_i}z_i^j+ \Op(\sqrt{m_i}), \;\textit{for
}i=1,2;\;j=1,\cdots,5,$$ }i=1,2;\;j=1,\cdots,5,$
where $z_i^j$'s are standard normal random variables and $\sumj z_i^j=0$ for $i=1,2$.
...
Note that, in general, by L'Hopital's rule, as $m_i\rightarrow \infty$,
$$\sqrt{m_i}\log\left(1+\frac{\sqrt{m_i}z_i^j+1/25}{m_ip^j} $\sqrt{m_i}\log\left(1+\frac{\sqrt{m_i}z_i^j+1/25}{m_ip^j} \right)=\frac{\log\left(1+\frac{\sqrt{m_i}z_i^j+1/25}{m_ip^j} \right)}{1/\sqrt{m_i}}\longrightarrow\frac{z_i^j}{p^j}, \;\textit{for
}i=1,2;\;j=1,\cdots,5.$$ }i=1,2;\;j=1,\cdots,5.$
Under the assumption that $m_1$ and $m_2$ are increasing at the same rate, let $m_1=m$ and $m_2=cm$, for some $c>0.$ Then as $m\rightarrow \infty$,