Files
@ 30d182b86860
Branch filter:
Location: AENC/switchchain/powerlaw_info.tex - annotation
30d182b86860
1.2 KiB
text/x-tex
Add better construction successrate measurement
42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d 42dadc398e7d | \documentclass{article}
\begin{document}
\section{Continuous powerlaw with minimum cut-off}
Exponent $\tau > 1$.
\subsection{Also maximum cut-off}
Let $m$ be the minimum, $M$ be the maximum.\\
PDF: $$f(x) = \frac{\tau-1}{m^{-(\tau-1)} - M^{-(\tau-1)}} x^{-\tau}$$
CDF: $$F(x) = \frac{m^{-(\tau-1)} - x^{-(\tau-1)}}{m^{-(\tau-1)} - M^{-(\tau-1)}}$$
Inverse: $$F^{-1}(y) = \left( (1-y) m^{-(\tau-1)} + y M^{-(\tau-1)} \right)^{\frac{-1}{\tau-1}}$$
i.e. linear interpolate between $M^{-(\tau-1)} < m^{-(\tau-1)}$.\\
$F^{-1}(0) = m$ and $F^{-1}(1) = M$.\\
For $m=1$ and $M$ steps of interpolation:\\
$F^{-1}(1/M) = \left( 1-M^{-1} + M^{-\tau} \right)^{\frac{-1}{\tau-1}}$\\
$F^{-1}(1-1/M) = \left( M^{-1} + M^{-(\tau-1)} - M^{-\tau} \right)^{\frac{-1}{\tau-1}}$
\subsection{No maximum cut-off}
For $M=\infty$ we have:\\
PDF: $$f(x) = \frac{\tau-1}{m^{-(\tau-1)}} x^{-\tau} = \frac{\tau-1}{m} \left(\frac{x}{m}\right)^{-\tau}$$
CDF: $$F(x) = \frac{m^{-(\tau-1)} - x^{-(\tau-1)}}{m^{-(\tau-1)}} = 1 - \left(\frac{x}{m}\right)^{-(\tau-1)}$$
Inverse: $$F^{-1}(y) = \left( (1-y) m^{-(\tau-1)} \right)^{\frac{-1}{\tau-1}} = m \; \left( 1-y \right)^{\frac{-1}{\tau-1}}$$
For interpolation:\\
$F^{-1}(0) = m$\\
$F^{-1}(1-\frac{1}{n}) = m\cdot n^{\frac{1}{\tau-1}}$
\end{document}
|