\documentclass{article}

%\VignetteEngine{knitr::knitr}
%\VignetteIndexEntry{The brief intro into the package}
%\VignetteDepends{}
%\VignetteKeywords{string, misc}
%\VignettePackage{pvar}

\usepackage[utf8]{inputenc}
 
<<setup, echo=FALSE, cache=FALSE>>=
library("pvar")
@



\title{pvar: the calculation and applications of $p$-variation }
\author{Vygantas Butkus \\
\small{Vygantas.Butkus@gmail.com}}
\date{\today}

\begin{document}
\maketitle

\begin{abstract}
In this vignette we will give a brief introduction to the $p$-variations.
Later on, we will give some illustrative examples of the main functions in
\emph{pvar} package.
\end{abstract}

\section{Introduction}

Let start with the definitions.
Originally $p$-variation is defined for a functions.
For a function $f:[0,1] \rightarrow R$ and $0 < p < \infty$ 
$p$-variation is defined as 
$$
 v_p(f) = \sup \left\{ \sum_{i=1}^m |f(t_i) - f(t_{i-1})|^p : 0=t_0<t_1<\dots<t_m=1, m \geq 1 \right\}.
$$

Suppose $X=\{X_{i}\}_{i=0}^{n}$ is any real-value sequence of numbers. 
We will call such sequence a \emph{sample}, 
whereas $n$ will be referred to as a \emph{sample size}.
The $p$-variation of the sample could be expressed as
$$
  v_p(X) = \max\left\{ \sum_{i=1}^k |X_{j_i} - X_{j_{i-1}}|^p :
  0=j_0<\dots<j_k=n,\; k=1,\dots,n  \right\}.
$$
The set of index $j_0,\dots,j_k$ that achieves the maximum is refereed as supreme partition.

Lots and detailed information on the known properties 
of $p$-variation can be found in the books of  Norvai\v{s}a and
Dudley (see \cite{DudleyNorvaisa1998} and \cite{DudleyNorvaisa1999}). 
There are formulated and proved a lot of the properties of $p$-variation.
In addition, the comprehensive bibliography of $p$-variation is include.


The core functions of the package is based on algorithm described in \cite{ButkusNorvaisa2018}.
The calculation of $p$-variation was inspired by paper \cite{NorvaisaRackauskas2008}, where
authors Norvai\v{s}a and Ra\v{c}kauskas presented
necessary and sufficient conditions
for the convergence in law of partial sum processes in
$p$-variation norm. Based on the result of this paper, Ra\v{c}kaskas proposed 
the method of testing sample of multiple shift in mean. This test is also implemented in this package.


The algorithm is written in C++, therefore it is quite fast.
It is worth noting, that \emph{pvar} package  could be used to calculate $p$-variation
for any piecewise monotone functions and to investigate
general properties of $p$-variation.



\section{Main function}

The main function in \emph{pvar} package is \emph{pvar} (that a surprise). It calculates the 
supreme partition and the value of $p$-variation. The main usage of \emph{pvar} function goes as follows
<<pvar_basic>>=
N <- 1000
p <- 2
x <- rwiener(N)
pv <- pvar(x, p)
pv
summary(pv)
@


Moreover, the data and supreme partition could be illustrated with \emph{plot} function
<<pvar_plot>>=
plot(pv)
@

The points in supreme partition (the red points) are actually used in $p$-variation. 
All the other points are redundant and could be dropped out with out any effect to $p$-variation.
This fact is illustrated in the example below.

<<pvar_partition>>=
pv.PP = pvar(x[pv$partition], TimeLabel=time(x)[pv$partition], 2)
plot(pv.PP, main='The same pvar without redundant points')
@

Moreover, $p$-variation on different intervals could be merged into unified sample. 
Let look at the example

<<pvar_add>>=
x = rwiener(1000)
pv1 = pvar(x[1:500], 2)
pv2 = pvar(x[500:length(x)], 2)
sum_pv1_pv2 = pv1 + pv2
@

The illustration of this looks like this

<<pvar_add_picture, echo=FALSE>>=
layout(matrix(c(1,3,2,3), 2, 2))
# par(mar=c(4,4,2,1))
plot(pv1, main="pv1")
plot(pv2, main="pv2")
plot(sum_pv1_pv2, main="p-variation of pv1+pv2")
layout(1)
@

\section{PvarBreakTest}

A. Ra\v{c}kauskas has proposed a method of detection of multiple changes in a sequence of independent observations.
This method is based on results presented in \cite{NorvaisaRackauskas2008}.
This method is realized in this package under the name of Pvar Break Test (or $PBT$ for short).
It is implemented in \emph{PvarBreakTest} function.

Lets consider the example. Let construct the data $x$ with multiple shifts in mean.

<<test_data, echo=TRUE>>=
set.seed(1)
MiuDiff <- 0.3
x <- rnorm(250*4, rep(c(0, MiuDiff, 0, MiuDiff), each=250))
@


<<test_plotdata, echo=FALSE>>=
plot(x, pch=19, cex=0.5, main="Original data, with several shifts in mean")
k <- 50
moveAvg <- filter(x, rep(1/k, k))
lines(time(x), moveAvg, lwd=2, col=2)
legend("topleft", c("sample", "moving average (k="%.%k%.%")"),
  lty=c(NA,1), lwd=c(NA, 2), col=1:2, pch=c(19,NA), pt.cex=c(0.7,1)
  ,inset = .03, bg="antiquewhite1")  
@

From the original data, it is not obvious if there are any breaks in sample.
To test whether there is a multiple break, we can apply \emph{PvarBreakTest} test.
The results are remarkable good. Look below.


<<test_result, echo=TRUE>>=
xtest <- PvarBreakTest(x)
xtest
@

<<test_plot, echo=FALSE>>=
plot(xtest)
@



%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 
\begin{thebibliography}{99}  

% Musu stripsnis
\bibitem{ButkusNorvaisa2018} 
V. Butkus, R. Norvai\v{s}a. Lith Math J (2018). https://doi.org/10.1007/s10986-018-9414-3

\bibitem{DudleyNorvaisa1998} 
R. M. Dudley, R. Norvai\v{s}a. An Introduction to
$p$-variation and Young Integrals, Cambridge, Mass., 1998.


\bibitem{DudleyNorvaisa1999} 
R. M. Dudley, R. Norvai\v{s}a. Differentiability of 
Six Operators on Nonsmooth Functions and $p$-variation, 
Springer Berlin Heidelberg, 
Print ISBN 978-3-540-65975-4, Lecture Notes in Mathematics Vol.~1703, 1999.

 
\bibitem{NorvaisaRackauskas2008}
R. Norvai\v{s}a, A. Ra\v{c}kauskas. Convergence in law of partial sum processes in p-variation norm // Lth. Math. J., Vol. 48, No. 2, 212–227, 2008.


\bibitem{Qian}
J. Qian. The $p$-variation of partial sum processes and the empirical process // Ph.D. thesis, Tufts University, 1997.




\end{thebibliography}



\end{document}