\documentclass[reqno]{amsart}
\usepackage{hyperref}
\usepackage{graphicx}
\AtBeginDocument{{\noindent\small
\emph{Electronic Journal of Differential Equations},
Vol. 2012 (2012), No. 11, pp. 1--8.\newline
ISSN: 1072-6691. URL: http://ejde.math.txstate.edu or http://ejde.math.unt.edu
\newline ftp ejde.math.txstate.edu}
\thanks{\copyright 2012 Texas State University - San Marcos.}
\vspace{9mm}}
\begin{document}
\title[\hfilneg EJDE-2012/11\hfil Cauchy-Kowalevski and polynomial ODEs]
{Cauchy-Kowalevski and polynomial ordinary differential equations}
\author[R. J. Thelwell, P. G. Warne, D. A. Warne\hfil EJDE-2012/11\hfilneg]
{Roger J. Thelwell, Paul G. Warne, Debra A. Warne} % in alphabetical order
\address{Roger J. Thelwell \newline
Department of Mathematics and Statistics,
James Madison University, Harrisonburg, VA 22807, USA}
\email{thelwerj@jmu.edu}
\address{Paul G. Warne \newline
Department of Mathematics and Statistics,
James Madison University, Harrisonburg, VA 22807, USA}
\email{warnepg@jmu.edu}
\address{Debra A. Warne \newline
Department of Mathematics and Statistics,
James Madison University, Harrisonburg, VA 22807, USA}
\email{warneda@jmu.edu}
\thanks{Submitted October 11, 2010. Published January 17, 2012.}
\subjclass[2000]{34A12, 34A34, 35A10}
\keywords{Automatic differentiation; power series; Taylor series;
\hfill\break\indent polynomial ODE; majorant; error bound}
\begin{abstract}
The Cauchy-Kowalevski Theorem is the foremost result guaranteeing
existence and uniqueness of local solutions for analytic
quasilinear partial differential equations with Cauchy initial data.
The techniques of Cauchy-Kowalevski may also be applied to
initial-value ordinary differential equations.
These techniques, when applied in the polynomial ordinary
differential equation setting, lead one naturally to a method
in which coefficients of the series solution are easily
computed in a recursive manner, and an explicit majorization
admits a clear a priori error bound.
The error bound depends only on immediately
observable quantities of the polynomial system; coefficients,
initial conditions, and polynomial degree.
The numerous benefits of the polynomial system are shown for
a specific example.
\end{abstract}
\maketitle
\allowdisplaybreaks
\section{Introduction}
The Cauchy-Kowalevski Theorem is the main tool in showing the
existence and uniqueness of local solutions for analytic quasilinear
partial differential equations (PDE) with Cauchy initial data.
Cauchy developed a proof in a restricted setting by 1842 \cite{Cauchy},
and in 1875 Kowalevski presented the full result \cite{Kowalevski};
existence of a unique solution to the general quasilinear system of
partial differential equations given initial conditions prescribed on
some non-characteristic curve. In \cite{Folland}, a proof in the fully
nonlinear setting is presented. The Cauchy-Kowalevski argument is based
on the construction of a power series solution, in which the coefficients
of the series expansion are reconstructed recursively, and the method
of majorants applied to verify that this solution converges locally.
Convergence is demonstrated by comparison with the analytic solution of
an associated PDE.
Although the Picard-Lindel\"of Theorem is the fundamental local
existence argument for a large class of initial value ordinary
differential equations (IVODE), in 1835 Cauchy demonstrated existence
and uniqueness in the ODE setting, applying a majorant based argument
similar to that both
he and Kowalevski would later use in the PDE setting.
That is, Cauchy methods can be used to show that $u$ satisfies
the real analytic ODE $d_t u(t) = f(u(t))$, where $u(0) = u_0$
using a constructive approach, provided $f(u)$ is analytic near $u_0$.
A nice treatment may be found in \cite{Driver}.
Given that the power series solution is directly accessible
via the Cauchy-Kowa\-levski construction but that the method is rarely
applied suggests practical difficulties.
In fact, the coefficients of the series solution can be tedious
to construct as typically posed,
as is a key constant in the comparison solution.
In this paper, we demonstrate that a subtle
recasting of the ODE system meliorates these difficulties:
the coefficients of the analytic solution become
remarkably easy to recover, and a computable choice of the
key constant in the majorization leads to an \textit{attractive a priori error bound.}
To make these ideas clear, we consider the simple quasilinear problem
\begin{equation}
\label{E:IVODE}
d_t u(t) = f(u(t)) := \frac{\mathrm{e}^{2 u(t)^2}}{\sin u(t)} ,
\quad\text{with } u(0) = 1.
\end{equation}
We first consider \eqref{E:IVODE} using the methods of Cauchy,
and identify steps in which the construction of solution becomes tedious.
We then recast the problem as a polynomial system, as might be done
when using Taylor series based automatic differentiation, and apply the
same methods. It will be clear the computations necessary to
generate the series solution are basic, and that a simple majorization
which depends only on the magnitude of the \textit{initial conditions}, the \textit{degree}
of the polynomial system and the magnitude of the \textit{constant coefficients}
of the system leads to an explicit bound of the remainder when approximating with the Taylor Polynomial.
Although not demonstrated here, the method applied is quite general.
The authors view this note as complimentary to \cite{Warne2006}.
Most importantly, we conjecture that it may be possible to extend the method to analytic IVPDE.
\section{Recasting (non)linear ODE(s) as polynomial systems: Why?}
Ordinary differential equations, particularly nonlinear and those with
singularities, play a fundamental role in understanding the principles
that govern the world around us.
Left in their original
(or \textit{classic}) form, various analytic and numeric methods exist which are
problem specific, yet there remains a need for a systematic method to calculate
solutions of general problems.
The approach presented here, perhaps first introduced by Cauchy and
subsequently rediscovered and coupled with power series
methods by Fehlberg in 1964 (\cite{Fehlberg1964}) and others since, is
simple and surprisingly general. A recasting of the original
ODE as a system of constant coefficient polynomial ODEs
via an introduction of auxiliary variables leads to a straight
forward iterative calculation of power series coefficients.
This allows a clear and systematic construction of numeric solutions and
provides an immediate and explicit \textit{a priori} error bound.
A polynomial system is useful computationally, and methods are
available to recast an impressively wide variety of ODEs into an augmented
polynomial system.
This includes ODEs with right-hand sides involving compositions
of exponential, logarithmic, and trigonometric functions,
as well as those involving the algebraic operations, including
exponentiation of complex power and encompassing general reciprocals and singularities.
For example, the second order ODE,
\begin{equation}
\label{deb}
y''\Big[ {1 + \frac{{\sqrt 2 }}
{{{{y'}^2}}}{{\big( {\frac{x}
{{yy'}}} \big)}^{\sqrt 2 - 1}}} \Big]
= \Big[ {\sqrt 2 {{\big( {\frac{x}
{{yy'}}} \big)}^{\sqrt 2 }} + 1} \Big]
\Big[ {\frac{y}{{{x^2}}} - \frac{{y'}}
{x}} \Big] + \frac{{{\pi ^2}}}{{16}}y ,
\end{equation}
with ${( {})' }: = \frac{d}{{dx}}$, used to
model the torsional deformation of a compressible elastic solid
cylinder composed of a generalized Blatz-Ko material, does not appear amiable to classic power
series methods, and yet its series solution to arbitrary order
is easily computable from its equivalent polynomial system \cite{Paullet}.
While such computational advantage is not the focus here,
see \cite{Barrio, Griewank,Jorba2005,Rall} for examples and discussion in
the automatic differentiation setting or
\cite{Parker1996,Pruett2003,Perlin1964,Stewart2009} in the ODE setting.
Recasting an ODE as a polynomial system also makes analysis tractable. We now explore
one such application: a proof of existence and uniqueness.
\section{Cauchy solution: the classic setting}
We begin with the precarious assumption that a locally analytic solution $u(t)$ to \eqref{E:IVODE} exists, and repeatedly differentiate the equation, using the fact that $f(u)$ is analytic in $u$ near the initial condition.
\begin{align*}
d_t^2 u(t) & = d_u f(u) d_t u \\
& = -{\frac {{{\rm e}^{4\,{u}^{2}}} \left( 4\,u\sin u -\cos
u \right) }{ \left( -1+ \cos^{2} u \right) \sin u }},
\notag \\
d_t^3 u(t) & = d_u^2 f(u) [d_t u]^2 + d_u f(u) d_t^2 u \notag \\
& = -{\frac {{{\rm e}^{6\,{u}
^{2}}} \left( 32\,{u}^{2}\cos^{2} u+2\, \cos^{2} u + 16\,u\sin u \cos u -5-32\,{u}^{2} \right) }{ \left( 1-2\, \cos^{2} u +
\cos^{4} u \right) \sin u }}
\\
d_t^4 u(t) & = d_u^3 f(u) [d_t u ]^3 + 3 d_u^2 f(u) d_t^2 u d_t u + d_u f(u) d_t^3 u \notag \\
& = {\frac {{{\rm e}^{8\,{u}^{2}}} \left( M_1 + M_2 \right) }{ \left( -1+3\, \cos^{2} u-3\, \cos^{4} u+ \cos^{6} u \right) \sin u }}
\end{align*}
where
\begin{gather*}
M_1 = -288\,{u}^{2} \cos^{3} u-22\, \cos^{3} u+384\,{u}^{3}
\sin u \cos^{2} u -140\,u\sin u \\
M_2 = 40\,u\sin u \cos^{2} u+37\,\cos u +288\,\cos u {u}^{2}
+384\,{u}^{3}\sin u
\end{gather*}
and
\begin{equation} \label{E:polynomial}
d_t^n u(t) = p_n(f(u),d_u f(u),d_u^2 f(u), \ldots , d_u^{n-1} f(u)),
\end{equation}
where $p_n(\cdot)$ denotes a polynomial in $n$ variables
(here taken from the set of derivatives of
$f$ with respect to $u$ of order less than $n$; i.e.,
$\{ d_u^{k-1} f \}, k=1,\ldots,n$, and having positive integer
coefficients).
By this process, all coefficients of the power series representation of
$u(t)$ may be built;
\begin{equation}
\label{E:PSS}
u(t) = \sum_{k=0}^\infty \frac{1}{k!} d_t^{k} u(0) \, t^k.
\end{equation}
Note that the form of the polynomial $p_n$ in expression
\eqref{E:polynomial}
allows the coefficients of the power series to be recovered
recursively, although the complexity of calculation may
(and usually does) grow exponentially.
By its very construction, this power series \eqref{E:PSS}
yields a unique classical solution to the initial-value ODE if it can be shown
to converge. Cauchy demonstrated convergence by comparison with
a related analytic initial-value ODE,
whose individual coefficients majorize (absolutely bound) those of
\eqref{E:PSS}. We briefly illustrate the argument.
We begin with the assumption of the theorem that $f(u)$ is analytic
in some interval of radius $R \in \mathbb{R}$ about $u=1$,
and remark that in practice $R$ might be quite difficult to determine.
Then for any positive $r