Documentation:Monte Carlo Equilibration

From ALPS
Revision as of 14:23, 10 September 2013 by Tamama (talk | contribs) (Synposis)

Jump to: navigation, search

Monte Carlo equilibration

Theory

We have a timeseries of N measurements obtained from a Monte Carlo simulation, i.e. y_0,y_1,\cdots,y_{N-1}.

Suppose \bar{y}_i = \beta_0 + \beta_1 x_i (s.t. i = 0, 1, \cdots, N-1) is the least-squares best fitted line, we attempt to minimize  S = \sum_i (y_i - \bar{y}_i)^2 w.r.t. \beta_0 and  \beta_1.

\frac{\partial S}{\partial \beta_0 } = 0 , \frac{\partial S}{\partial \beta_1 } = 0  :


\Rightarrow \left( 
\begin{array}{cc} 
N               & \sum_i x_i     \\
\sum_i x_i & \sum_i x_i^2
\end{array}
\right)
\left(
\begin{array}{c}
\beta_0 \\ 
\beta_1
\end{array}
\right) =
\left(
\begin{array}{c}
\sum_i y_i  \\ 
\sum_i x_i y_i 
\end{array}
\right)


\Rightarrow \left(
\begin{array}{c}
\beta_0 \\ 
\beta_1
\end{array}
\right) =
\frac{1}{N\sum_i x_i^2 - (\sum_i x_i)^2}
\left( 
\begin{array}{cc} 
\sum_i x_i^2 & -\sum_i x_i     \\
-\sum_i x_i     & N
\end{array}
\right)
\left(
\begin{array}{c}
\sum_i y_i  \\ 
\sum_i x_i y_i 
\end{array}
\right)


\Rightarrow \beta_1 = \frac{N \sum_i x_i y_i - \sum_i x_i \sum_i y_i}{N\sum_i x_i^2 - (\sum_i x_i)^2}


Slope of best-fitted line


\Rightarrow \beta_1 = \frac{\sum_i (x_i - \bar{x}_i)( y_i - \bar{y}_i) }{\sum_i (x_i - \bar{x}_i)^2}  \,\,\,\,\, \left( = \frac{s_{xy}}{s_{xx}} \right)


Error in slope of best-fitted line


\mathrm{Var}({\beta_1}) 
= \mathrm{Var} \left( \frac{\sum_i (x_i - \bar{x}_i)( y_i - \bar{y}_i) }{\sum_i (x_i - \bar{x}_i)^2} \right)
= \frac{\sum_i (x_i - \bar{x}_i)^2 \mathrm{Var}(y_i) }{(\sum_i (x_i - \bar{x}_i)^2)^2}


Denoting \mathrm{Var}(y_i) = \sigma_{y}^2 , we have:

 \Rightarrow \mathrm{Var}({\beta_1}) = \sigma_{\beta_1}^2 = \frac{\sigma_y^2}{\sum_i (x_i - \bar{x}_i)^2}

 \Rightarrow \mathrm{Var}({\beta_1}) = \sigma_{\beta_1}^2 = \frac{ 12 \, \sigma_y^2 }{ N(N^2 - 1) }


Hypothesis testing


 
\begin{array}{l}
H_0 : \beta_1 = 0 \\
H_1 : \beta_1 \ne 0 
\end{array}


Using the standard z-test, we reject H_0 at confidence interval \alpha if


\left| \frac{\beta_1 - 0 }{ \sigma_{\beta_1} } \right| > z_{\frac{1-\alpha}{2}}


Practice

Synposis

pyalps.checkNonSteadyState(outfile, observables, confidenceInterval=0.63, simplified=False, includeLog=False)
argument default type remark
outfile -- Python str ALPS hdf5 output file(name)
confidenceInterval 0.01  \mathrm{tolerance} = \frac{X^\mathrm{(fit)} (t_\mathrm{final}) - X^\mathrm{(fit)} (t_\mathrm{initial})}{\bar{X}}
simplified False shall we combine the checks of all observables as 1 final boolean answer?
includeLog False shall we print the detailed log?