Time-Dependent Symmetries of Variable-Coefficient Evolution Equations and Graded Lie Algebras
Extracted source text ยท arXiv solv-int/9705014
\documentstyle[12pt]{article}
\setlength{\textwidth}{160mm}
\setlength{\textheight}{230mm}
\setlength{\oddsidemargin}{5mm}
\setlength{\topmargin}{-20mm}
%\setlength{\parindent}{0pt}
\setlength{\parskip}{6pt plus 2pt minus 1 pt}
\frenchspacing
\begin{document}
\centerline
{{\large {\bf Time-Dependent Symmetries of Variable-Coefficient}}}
\centerline{{\large {\bf Evolution Equations and Graded Lie Algebras}}}
\vskip 2mm
\centerline {W. X. Ma\footnote{Current address: Dept. of Math., City
University of Hong Kong, Kowloon, Hong Kong; Email:
mawx@math.cityu.edu.hk},
R. K. Bullough\footnote{Email: Robin.Bullough@umist.ac.uk},
P. J. Caudrey\footnote{Email: Philip.Caudrey@umist.ac.uk}
and W. I. Fushchych\footnote{Email: math@nonlinear.kiev.ua}
}
\centerline{Department of Mathematics,
University of Manchester}
\centerline{Institute of Science and Technology,
Manchester M60
1QD, UK}
\newtheorem{thm}{Theorem}%[section]
\newtheorem{lemma}{Lemma}%[section]
\newcommand{\R}{\mbox{\rm I \hspace{-0.9em} R}}
\newcommand{\Z}{\mbox{\rm Z}}
\def\be{\begin{equation}}
\def\ee{\end{equation}}
\def\ba{\begin{array}}
\def\ea{\end{array}}
\def\bea{\begin{eqnarray}}
\def\eea{\end{eqnarray}}
\def\la {\lambda}
\def \part {\partial}
\def \al {\alpha}
\def \de {\delta}
\setlength{\baselineskip}{18pt}
\begin{abstract}
Polynomial-in-time dependent symmetries are analysed for
polynomial-in-time dependent evolution equations.
Graded Lie algebras, especially Virasoro algebras, are used to construct
nonlinear variable-coefficient evolution equations,
both in $1+1$ dimensions and in $2+1$ dimensions,
which possess higher-degree polynomial-in-time
dependent symmetries. The theory
also provides a kind of new realisation of graded Lie algebras. Some
illustrative examples are given.
\end{abstract}
It is well known that the usual family of KdV equations has
polynomial-in-time dependent symmetries (ptd-symmetries)
which are only of the first-degree.
This is because only master symmetries
of first degree are so far found.
Moreover there are usually
\footnote{
%However, there are some exceptions, for example,
The Benjamin-Ono equation is a counter-example.}
no higher-degree ptd-symmetries
for time-independent integrable equations in $1+1$ dimensions;
but this may not be so in $2+1$ dimensions.
However a form of special graded Lie algebras, namely
centreless Virasoro symmetry algebras is apparently common
to all time-independent
integrable equations in whatever dimensions both in the continuous case and
in the discrete case. This feature would therefore seem to be
an important one in the discussion
of integrability and integrable nonlinear equations.
For the higher dimensional integrable equations,
there may also exist still more general graded symmetry Lie algebras.
The purpose of the present paper
is to discuss ptd-symmetries for evolution equations
with polynomial-in-time dependent
coefficients (conveniently expressed in terms of monomials in $t$ as in eqn.
(\ref{pee}) below).
We provide a purely algebraic structure for constructing such
integrable equations with these forms of symmetries.
This way we show there do exist
integrable equations in $1+1$ dimensions
which possess these forms of symmetries and we construct actual examples.
Graded Lie algebras, and especially centreless Virasoro algebras, are used
for these constructions. In consequence new features are
extracted from the graded Lie algebras which provide
new realisations of
these algebras and most particularly of the centreless Virasoro algebras.
We first define a symmetry for an evolution equation, linear and nonlinear
\cite{Olver} \cite{Fuchssteiner1983} \cite{ChenL} \cite{MaF} \cite{ChengLB}.
For a given evolution equation
$u_t=K(u)$, a vector field $\sigma (u)$ is called its symmetry if
$\sigma (u)$ satisfies its linearized equation
\be \frac {d \sigma (u)}{dt}=K'[\sigma ],\ {\rm i.e.} \ \frac {\part \sigma}
{\part t}=[K,\sigma ],\label{sd}\ee
where the prime and $[\cdot ,\cdot]$ denote the Gateaux derivative and
the Lie product
\be K'[S]=\frac{\part }{\part \varepsilon }K(u+\varepsilon S)
|_{\varepsilon =0},\ [K,\sigma ]=K'[\sigma ]-\sigma '[K],\ee
respectively.
Of course, a symmetry $\sigma$
may also depend explicitly on the time variable $t$.
For example, $\sigma $ may be of
polynomial type in $t$, i.e.
\be \sigma (t,u)=\sum_{j=0}^{n}
\frac{t^j}{j!}S_j(u)=S_0+tS_1+\cdots +\frac {t^n}{n!}S_n,
\label{ps} \ee
where the vector fields $S_j(u),\ 0\le j\le n,$ do not depend explicitly
on the time variable $t$.
If we consider a variable-coefficient
evolution equation $u_t=K(t,u)$ of the form
\be u_t=K(t,u)=\sum_{i=0}^m \frac {t^i}{i!} T_i(u)=T_0+tT_1+\cdots +
\frac{t^m}{m!} T_m,\label{pee}\ee
where the vector fields
$T_i(u),\ 0\le i\le m,$ do not depend explicitly on the time variable $t$,
either, then a precise result may be obtained which states
(\ref{ps}) is a symmetry of (\ref{pee}).
At this stage, we can have
\bea
\frac {\part \sigma }{\part t} & =& \sum_{i=0}^n\frac{t^{i-1}}{(i-1)!}
S_i(u)=\sum_{k=0}^{n-1}\frac{t^k}{k!}S_{k+1}(u),\nonumber \\
\left [K,\sigma \right] &= & \Bigl [\sum_{i=0}^m \frac{t^i}{i!}T_i(u),
\sum_{j=0}^n \frac{t^j}{j!}S_j(u)\Bigr ]\nonumber \\
&=& \sum_{k=0}^{m+n} \frac{t^k}{k!}
\sum_{\ba {c} \vspace{-6mm}\\{_{i+j=k}} \vspace{-2mm} \\ {_{0\le i\le m}}
\vspace{-2mm} \\ {_{0\le j\le n}} \ea }
{k \choose i}
[T_i,S_{j}] .\nonumber
\eea
Therefore a simple comparison of each power of $t$ in (\ref{sd}) leads to
\bea
&&S_{k+1}=
\sum_{\ba {c} \vspace{-6mm}\\
{_{i+j=k}} \vspace{-2mm} \\ {_{ 0\le i\le m}}
\vspace{-2mm} \\ {_{0\le j\le n}} \ea }
{k \choose i} [T_i,S_{j}], \ 0\le k\le n-1, \label{grule1}\\
&&
\sum_{\ba {c} \vspace{-6mm}\\
{_{ i+j=k}} \vspace{-2mm} \\ {_{ 0\le i\le m}}
\vspace{-2mm} \\ {_{0\le j\le n}} \ea }
{k \choose i} [T_i,S_{j}]=0, \ n\le k\le m+n.
\label{grule2}\eea
These equalities in (\ref{grule1}) and (\ref{grule2})
constitute a necessary and sufficient condition to state that (\ref{ps})
is a symmetry of (\ref{pee}). If we look at them a little more,
it may be seen that
\bea
&&S_1=[T_0,S_0],\nonumber \\
&&S_2=[T_0,S_1]+[T_1,S_0],\nonumber \\
&& \cdots\cdots \nonumber \\
&&S_n={n-1 \choose 0}[T_0,S_{n-1}]+{n-1 \choose 1}[T_1,S_{n-2}]+\cdots +
{n-1 \choose n-1}[T_{n-1}, S_0],\nonumber
\eea
where $T_i=0,\ i\ge m+1$, and so
a higher-degree ptd-symmetry
$\sigma (t,u)$ defined by (\ref{ps}) is
determined completely by a vector field $S _0$.
However this vector field $S _0$ needs to satisfy (\ref{grule2}).
This kind of vector field $S_0$ is a generalisation of the
master symmetries defined in \cite{Fuchssteiner1983}
which here we still call a master symmetry of
degree $n$ for the more general evolution equation, eqn. (\ref{pee}).
We conclude the discussion above as a theorem.
\begin{thm} \label{tpds}
Let $\rho$ be a vector field not depending explicitly on the
time variable $t$. Define
\be S_0(\rho) = \rho ,\
S_{k+1}(\rho) = \sum_{j=0}^k {k \choose j} [T_j,S_{k-j}(\rho)],
\ k\ge 0, \label{defs_{k+1}}
\ee
where we assume $T_i=0,\ i\ge m+1$.
If there exists $n\in N$ so that $S_j(\rho)=0,\ j\ge n+1$, then
\be \sigma (\rho) =\sum_{j=0}^n\frac {t^j}{j!}S_j(\rho)\label{defsigmaofrho}
\ee
is a polynomial-in-time dependent symmetry of the evolution equation,
eqn. (\ref{pee}).
\end{thm}
We shall go on to construct variable-coefficient
integrable equations which possess
higher-degree ptd-symmetries as defined by (\ref{ps}).
We need to start from the centreless Virasoro algebra
\be \left \{ \ba {l}
\left [K_{l_1},K_{l_2}\right]=0,\ l_1,l_2\ge 0,
\vspace{2mm} \\
\left[K_{l_1},\rho_{l_2}\right]=(l_1+\gamma )K_{l_1+l_2},\ l_1,l_2\ge 0,
\vspace{2mm} \\
\left [\rho_{l_1},\rho_{l_2}\right]=(l_1-l_2)\rho_{l_1+l_2},
\ l_1,l_2\ge 0, \ea
\right. \label{Va}
\ee
in which
the vector fields $K_{l_1}=K_{l_1}(u),\ \rho_{l_2}=\rho_{l_2}(u),
\ l_1,l_2\ge 0$, do not
depend explicitly on the time variable $t$ and $\gamma $ is a fixed constant.
Although the vector fields $\rho_l,\ l\ge 0,$ are not symmetries of
any equations that we want to discuss, an algebra isomorphic to
this kind of Lie algebra
commonly arises as a symmetry algebra for many well-known continuous and
discrete integrable equations \cite{ChenL} \cite{MaF} \cite{ChengLB}.
In eqn. (\ref{Va}), the vector fields $\rho_l,\ l\ge 0,$
may provide the generators of
Galilean invariance \cite{Fushchych} and
invariance under scale transformations
for any standard equation $u_t=K_k(u)$.
Let us choose a set of specific vector fields
\be T_j=K_{i_j}, \ 0\le j\le m,\label{Tchoose}\ee
which yields the following variable-coefficient evolution equation
\be u_t=K_{i_0}+tK_{i_1}+\frac{t^2}{2!}K_{i_2}+\cdots +\frac{t^m}{m!}
K_{i_m}.\label{spee}\ee
This equation
still has a hierarchy of time-independent symmetries $K_l, \ l\ge 0,$
and therefore it is integrable in the sense of symmetries \cite{Fokas}.
What is more, it
will inherit many integrable properties of $u_t=K_l,\ l\ge0$.
For example, if $u_t=K_l,\ l\ge 0,$ have Hamiltonian structures of the form
\[ u_t=K_l=J\frac {\delta H_l}{\delta u}, \ l\ge 0,\]
where $J$ is a symplectic operator and
$H_l,\ l\ge 0,$ do not depend explicitly on $t$,
then the $H_l$
are still conserved densities of eqn. (\ref{spee}) and eqn. (\ref{spee})
is then completely integrable in the commonly used sense for pdes.
In what follows, we need to prove
that $\rho_l$ is a master symmetry (as explained above) of
degree $m+1$ of eqn. (\ref{spee}).
%Thus we need to show that
%$S_j(\rho_l)=0,\ j\ge m+2$, where the $S_j(\rho_l)$ are
%defined by (\ref{defs_{k+1}}).
In fact, according to (\ref{defs_{k+1}}),
we have
\[ S_0(\rho_l)=\rho_l,\
S_{k+1}(\rho_l)=[T_k,S_0(\rho_l)]=[K_{i_k},\rho_l]
=(i_k+\gamma )K_{i_k+l},
\ 0\le k\le m,
\]
and further we can prove that $S_j(\rho_l)=0$ when $j\ge m+2$,
which shows that
$\rho_l$ is a master symmetry of
degree $m+1$ of eqn. (\ref{spee}).
Therefore we obtain a hierarchy of ptd-symmetries of the form
\bea \sigma _l(t,u)&=&\sum_{j=0}^{m+1}\frac{t^j}{j!} S_j(\rho_l)=
\sum_{j=1}^{m+1}\frac{i_{j-1}+\gamma }{j!}t^jK_{i_{j-1}+l}+\rho_l\nonumber\\
&=& \sum_{j=0}^{m}\frac{i_{j}+\gamma }
{(j+1)!}t^{j+1}K_{i_{j}+l}+\rho_l,\ l\ge 0,
\label{sps}
\eea
for the variable-coefficient and integrable equation (\ref{spee}).
Moreover these higher-degree ptd-symmetries
together with time-independent symmetries $K_l,\ l\ge 0$,
constitute the same centreless Virasoro algebra as
(\ref{Va}), namely
\be \left \{ \ba {l}
\left [K_{l_1},K_{l_2}\right]=0,\ l_1,l_2\ge 0,
\vspace{2mm} \\
\left[K_{l_1},\sigma_{l_2}\right]=(l_1+\gamma )K_{l_1+l_2},\ l_1,l_2\ge 0,
\vspace{2mm} \\
\left[\sigma_{l_1},\sigma_{l_2}\right]=(l_1-l_2)\sigma_{l_1+l_2},
\ l_1,l_2\ge 0.\ea
\right. \label{hVa}
\ee
For example, we can calculate that
\bea \left [\sigma_{l_1},\sigma_{l_2}\right] & = &
\Bigl[\sum_{j=0}^m\frac{i_j+\gamma }{(j+1)!}t^{j+1}K_{i_j+l_1}+\rho_{l_1},
\sum_{j=0}^m\frac{i_j+\gamma }{(j+1)!}t^{j+1}K_{i_j+l_2}+\rho_{l_2}\Bigr]
\nonumber\\
&=&
\Bigl[\sum_{j=0}^m\frac{i_j+\gamma }
{(j+1)!}t^{j+1}K_{i_j+l_1},\rho_{l_2}\Bigr]
+ \Bigl[\rho_{l_1},
\sum_{j=0}^m\frac{i_j+\gamma }{(j+1)!}t^{j+1}K_{i_j+l_2}\Bigr]
+\left[\rho_{l_1},\rho_{l_2}\right]
\nonumber \\
&=& \sum_{j=0}^m\frac{(l_1-l_2)(i_j+\gamma )}{(j+1)!}t^{j+1}K_{i_j+l_1+l_2}
+(l_1-l_2)\rho_{l_1+l_2}\nonumber\\
&& = (l_1-l_2)\sigma_{l_1+l_2}. \nonumber
\eea
The algebra (\ref{hVa}) also gives us
a new realisation of centreless Virasoro algebras.
By now we may very much see that there exist
higher-degree ptd-symmetries for some
evolution equations in $1+1$ dimensions. Moreover our derivation
does not refer to any particular choices of dimensions and space variables.
Hence the evolution equation (\ref{spee}) may be not only both
continuous and discrete, but also both $1+1$ and $2+1$ dimensional.
Actually there are many
integrable equations which possess a centreless Virasoro
algebra (\ref{Va}) (see \cite{ChenL} \cite{MaF} \cite{ChengLB}
\cite{Ma1990} \cite{OevelFZ}
for example).
Among the most famous examples are the KdV hierarchy
in the continuous case and the Toda lattice hierarchy
in the discrete case.
Through the theory above, we can say that
a KdV-type equation
\be u_t=tK_0+K_1=tu_x+u_{xxx}+6uu_x\ee
possesses a hierarchy of second-degree time-polynomial-dependent symmetries
\be
\sigma _l=\frac 32 tK_{l+1}+\frac 14t^2K_{l}+\rho_l,\ l\ge 0,
\ee
where the vector fields $K_l,\,\sigma _l,\ l\ge0,$ are defined by
\[ K_l=\Phi^l u_x,\ \rho_{l}=\Phi ^l(u+\frac 12 xu_x), \ \Phi = \part ^2+
4u+2u_x\part ^{-1},\ l\ge 0.\]
They constitute a centreless Virasoro algebra (\ref{Va}) with $\gamma =
\frac 12 $ \cite{Ma1990} \cite{Ma1992} and thus so do the symmetries
$K_l,\, \sigma_l,\ l\ge 0$.
We can also conclude that a Toda-type lattice equation
\bea && (u(n))_t=\left(\ba {c} p(n) \vspace{2mm} \\ v(n) \ea \right) _t
=K_0+tK_1+\frac{t^2}{2!}K_0\nonumber\\
&& =(1+\frac 12 t^2)\left(\ba {c} v(n)-v(n-1) \vspace{2mm}\\
v(n)(p(n)-p(n-1))\ea \right)\nonumber \\
&& \quad
+ t\left(\ba {c} p(n)(v(n)-v(n-1))+v(n)(p(n+1)-p(n-1)) \vspace{2mm}\\
v(n)(v(n-1)-v(n+1))+v(n)(p(n)^2-p(n-1)^2)\ea \right)
\eea
possesses a hierarchy of third-degree time-polynomial-dependent symmetries
\be
\sigma _l=tK_l+t^2K_{l+1}+\frac 16 t^3 K_l+\rho _l, \ l\ge 0,\ee
where the corresponding vector fields are defined by
\bea &&
K_l=\Phi ^{l}K_0,\ K_0=\left ( \ba {c} v-v^{(1)}\vspace{1mm}
\\ v(p-p^{(-1)})\ea \right ),
\ l\ge0 , \nonumber
\\ && \rho_{l}=\Phi ^l \rho_0,
\ \rho_0=\left ( \ba {c} p\vspace{2mm}
\\ 2v\ea \right ), \ l\ge 0, \nonumber
\eea
in which the hereditary operator $\Phi$ is defined by
\[ \Phi=\left ( \ba {cc} p&(v^{(1)}E^2-v)(E-1)^{-1}v^{-1}\vspace{1mm}
\\ v(E^{-1}+1)& v(pE-p^{(-1)})(E-1)^{-1}v^{-1}\ea \right ).
\]
Here we have used a normal shift operator $E$: $(Eu)(n)=u(n+1)$ and
$u^{(m)}=E^mu$.
These discrete
vector fields $K_l,\ l\ge 0$, (see \cite{Tu} for more information)
together with the discrete vector fields $\rho _l,\ l\ge 0$,
constitute a centreless Virasoro algebra (\ref{Va})
with $\gamma =1$ \cite{MaF} and the symmetry
Lie algebra of $\sigma _l,\
l\ge 0$ and $K_l,\ l\ge 0$, has the same commutation relations as that
Virasoro algebra.
More generally, we can consider further algebraic structures by starting
from a more general graded Lie algebra.
In keeping with the notation in \cite{Ma1992a},
let us write a graded Lie algebra consisting
of vector fields not depending explicitly
on the time variable $t$ as follows:
\be E(R)=\sum_{i=0}^\infty E(R_i),\ [E(R_i), E(R_j)]\subseteq E(R_{i+j-1}),
\ i,j\ge 0,\label{gradeda}\ee
where $E(R_{-1})=0$. Note that such a graded Lie algebra is called a master
Lie algebra in \cite{Ma1992a} since it is actually not
a graded Lie algebra as defined in \cite{Kac}.
However we still call it a graded Lie algebra because it is very similar.
Choose
\be T_i=K_i\in E(R_0) ,\ 0\le i\le m,\label{Tchoose2}\ee
and consider a
variable-coefficient evolution equation
\be u_t=\sum_{i=0}^m \frac {t^i}{i!}T_i=K_0+tK_1+\frac {t^2}{2!}K_2+\cdots
+\frac{t^m}{m!}K_m.\label{spee2}\ee
Before we state the main result, we derive two properties of
the generating vector fields $S_j,\ j\ge 0.$
\begin{lemma}
Assume that $T_i,\ 0\le i\le m,$ are defined by (\ref{Tchoose2}),
and let $l\ge 0$ and $\rho_l\in E(R_l)$. Then the vector fields $S_j(\rho_l),
\ j\ge 0$, defined by (\ref{defs_{k+1}}) satisfy the following property
\bea && S_{(\al -1)(m+1)+\beta }(\rho_l)\in \sum_{i=0}^{l-\al }E(R_i),
\ 1\le \al \le l,\ 1\le \beta \le m+1, \label{sigmaprop1}\\
&& S_j(\rho_l)=0,\ j\ge l(m+1)+1. \label{sigmaprop2}
\eea \label{sigmaproplemma}
\end{lemma}
{\bf Proof:}
Note the definition (\ref{defs_{k+1}}) of $S_{j}(\rho_l),\ j\ge 0$, and
$T_i=K_i,\ 0\le i\le m$.
We can calculate that
\bea && S_{\al (m+1)+\beta +1}(\rho_l)=\sum_{\gamma=0 }^m
{ \al (m+1)+\beta \choose \gamma}
\left [K_{\gamma },S_{\al (m+1)+\beta
-\gamma}(\rho_l) \right ]\nonumber \\
&& =\sum_{\gamma =0}^{\beta -1}
{\al (m+1)+\beta \choose \gamma }
\left [K_{\gamma },S_{\al (m+1)
+\beta -\gamma}(\rho_l) \right ]\nonumber \\
&&\quad +\sum_{\gamma =\beta }^m
{ \al (m+1)+\beta \choose \gamma}
\left [K_{\gamma },S_{(\al -1) (m+1)
+[(m+1)-(\gamma-\beta )]}(\rho_l) \right ]\nonumber \\
&& \in \sum_{i=0}^{l-(\al +2)} E(R_i)+\sum_{i=0}^{l-(\al +1)}
E(R_i)=\sum_{i=0}^{l-(\al +1)} E(R_i),\nonumber
\eea
where in the last but one step we have used the induction assumption.
This result shows that (\ref{sigmaprop1}) is true by mathematical induction.
The proof of
(\ref{sigmaprop2}) is the same so that the proof of the Lemma is complete.
$\vrule height 3mm width 1mm depth 0.4mm$
\begin{lemma} \label{lielemma}
Assume that $T_i,\ 0\le i\le m,$ are defined by (\ref{Tchoose2}),
and let $l_1,l_2\ge 0$ and $\rho_{l_1}\in E(R_{l_1}),\
\rho_{l_2}\in E(R_{l_2})$. Then we have
\be S_k([\rho_{l_1},\rho_{l_2}])=\sum_{i+j=k}{k\choose i}
[S_i(\rho_{l_1}),S_j(\rho_{l_2})],\ k\ge 0,\label{s_kprodp}\ee
where the $S_j(\rho),\ j\ge 0$, are defined by (\ref{defs_{k+1}}).
\end{lemma}
{\bf Proof:}
We use mathematical induction to prove the required result.
Noting that $T_i=K_i,\ 0\le i\le m,$ we can calculate that
\bea && S_{k+1}([\rho_{l_1},\rho_{l_2}])= \sum_{i+j=k}{k\choose i}\Bigl[
K_i,S_j([\rho_{l_1},\rho_{l_2}])\Bigr] \nonumber\\
&&= \sum_{i+j=k}{k\choose i}\Bigl[
K_i,\sum_{\al +\beta =j}{j \choose \al }\left[S_\al (\rho_{l_1}),
S_\beta (\rho_{l_2})\right]\Bigr]
\ \quad {\rm (by\ the \ induction\ assumption)}\nonumber\\
&&= \sum_{i+j=k}{k\choose i}\sum_{\al +\beta =j}{j \choose \al }
\Bigl[K_i, \left[S_\al (\rho_{l_1}),
S_\beta (\rho_{l_2})\right]\Bigr] \nonumber\\
&&= \sum_{i+\al +\beta =k}\frac {k!}{i!\al !\beta !}\Bigl[
K_i,\left[S_\al (\rho_{l_1}),
S_\beta (\rho_{l_2})\right]\Bigr] \nonumber\\
&&= \sum_{i+\al +\beta =k}\frac {k!}{i!\al !\beta !}\left \{\Bigl[
\left[K_i,S_\al (\rho_{l_1})\right],
S_\beta (\rho_{l_2})\Bigr]+ \Bigl [ S_\al (\rho_{l_1}),\left [ K_i, S_\beta
(\rho_{l_2})\right] \Bigr ]\right \}\nonumber \\
&&= \sum_{j +\beta =k}{k\choose j} \Bigl[
\sum_{i+\al =j}{j\choose i}\left [K_i,S_\al (\rho_{l_1})\right],
S_\beta (\rho_{l_2})\Bigr] \nonumber\\
&& \qquad +\sum_{\al +j=k}{k \choose j }\Bigl [S_\al (\rho_{l_1}),
\sum_{i+\beta =j}{j\choose i}\left[K_i,
S_\beta (\rho_{l_2})\right]\Bigr] \nonumber\\
&& =\sum_{j+\beta =k}{k\choose j} \left [S_{j+1}(\rho_{l_1}),S_{\beta}
(\rho_{l_2})\right]+
\sum_{\al +j =k}{k \choose j } \left [S_{\al }(\rho_{l_1}),S_{j+1}
(\rho_{l_2})\right]\nonumber \\
&& =\sum_{i+j=k+1}{k+1\choose i} \left[ S_i(\rho_{l_1}), S_j(\rho_{l_2})
\right],\ k\ge 0,\nonumber
\eea
and this yields the key step in the mathematical induction. On the other
hand, we easily see that
\[ S_0([\rho_{l_1},\rho_{l_2}])=[\rho_{l_1},\rho_{l_2}]
=[S_0(\rho_{l_1}),S_0(\rho_{l_2})].\]
Therefore mathematical induction gives the proof of the equality
(\ref{s_kprodp}).
$\vrule height 3mm width 1mm depth 0.4mm$
\begin{thm}
Assume that $T_i,\ 0\le i\le m,$ are defined by (\ref{Tchoose2}),
and let $l\ge 0$ and $\rho_l\in E(R_l)$. Then the vector field
\be \sigma (\rho_l)=\sum_{j=0}^{l(m+1)}\frac{t^j}{j!}S_j(\rho_l),
\label{sigmarhol}\ee
where the
$S_j(\rho_l), \ 0\le j\le l(m+1)$, are defined by (\ref{defs_{k+1}}),
is a time-independent symmetry of (\ref{spee2}) when $l=0$ and
a polynomial-in-time dependent symmetry of (\ref{spee2}) when $l>0$.
Furthermore we have
\be [\sigma (\rho_{l_1}),
\sigma (\rho_{l_2})]=\sigma ([\rho_{l_1},\rho_{l_1}]),\ \rho_{l_1}\in
E(R_{l_1}),\ \rho_{l_2}\in E(R_{l_2}), \ l_1,l_2\ge 0, \label{liep}
\ee
and thus all symmetries $\sigma (\rho_l)$ with $\rho_l\in E(R_l),\ l\ge 0$,
constitute the same graded Lie algebra as (\ref{gradeda}) and
the map $\sigma: \rho_l\mapsto \sigma(\rho_l)$ is a Lie homomorphism
between two graded Lie algebras $E(R)$ and $\sigma(E(R))$.
\end{thm}
{\bf Proof:} By Lemma \ref{sigmaproplemma}, we can observe that
$\sigma (\rho_l)$ defined by (\ref{sigmarhol}) is a symmetry of
(\ref{spee2}). We go on to prove (\ref{liep}). Assume that
$\rho_{l_1}\in E(R_{l_1}),\ \rho_{l_2}\in E(R_{l_2}),\ l_1,l_2\ge 0$.
By Lemma \ref{sigmaproplemma}
and Lemma \ref{lielemma}, we can make the following
calculation
\bea && \left [\sigma (\rho_{l_1}),\sigma (\rho_{l_2}) \right]
=\Bigl [\sum_{i=0}^{l_1(m+1)}\frac {t^i}{i!}S_i(\rho_{l_1}),
\sum_{j=0}^{l_2(m+1)}\frac {t^j}{j!}S_j(\rho_{l_2})\Bigr]
\nonumber \\
&& =\sum_{k=0}^{(l_1+l_2-1)(m+1)}\frac {t^k}{k!}\sum_{i+j=k}
{k \choose i}\left[ S_i(\rho_{l_1}), S_j(\rho_{l_2})\right] \ \
\ {\rm (by\ Lemma\ \ref{sigmaproplemma})}\nonumber\\
&& =\sum_{k=0}^{(l_1+l_2-1)(m+1)}\frac{t^k}{k!} S_k([\rho_{l_1},\rho_{l_2}])
\ \ \ {\rm (by\ Lemma\ \ref{lielemma})}\nonumber \\
&& = \sigma ([\rho_{l_1},\rho_{l_2}]).\nonumber
\eea
The rest is then obvious and the required result is obtained.
$\vrule width 1mm height 3mm depth 0.4mm$
A graded Lie algebra has been exhibited for the time-independent
KP hierarchy \cite{OevelF}
in \cite{Fuchssteiner1983}
\cite{Ma1992a}, and it includes a centreless Virasoro algebra
\cite{ChengLB} \cite{ChenLL}.
The ordinary time-independent KP equation being considered here
is the following
\be u_t=\part ^{-1}_xu_{yy}-u_{xxx}-6uu_x .\label{KP1}\ee
From this we may now go on to generate the corresponding
graded Lie algebra of ptd-symmetries
for a resulting new set of variable-coefficient KP equations, but in this
connection the reader must be referred to the comparable analysis
in \cite{MaBC} mentioned below.
The idea of using graded Lie algebras as described in this paper
is rather similar to the thinking
used to extend the inverse scattering transform
from $1+1$ to higher dimensions
\cite{Caudrey}. Moreover the resulting symmetry algebra
consisting of the $\sigma(\rho_l), l\ge 0$, provides a new realisation
of a graded Lie algebra (\ref{gradeda}). The theory
also shows us that more information can be extracted from
graded Lie algebras, which is itself very interesting. What is more,
we have shown here that there do exist various integrable equations
in $1+1$ dimensions, such as KdV-type equations,
possessing higher-degree polynomial-in-time dependent symmetries.
We report a graded Lie algebra of ptd symmetries for a corresponding
new set of variable coefficient {\it modified} KP equations in a second
article \cite{MaBC}. In \cite{MaBC} we display this modified KP
hierarchy explicitly, the time independent modified KP equation being,
in comparison with (\ref{KP1}), the equation
\be u_t=\frac14 u_{xxx}-\frac 38 u^2u_x-\frac 34 u_x\part ^{-1}_xu_y+\frac 34
\part ^{-1}_x u_{yy} .\label{KP2}\ee
In \cite{MaBC} we show also that this hierarchy actually has {\it two}
Virasoro algebras and {\it two} graded Lie algebras.
We also hope to show elsewhere the connections between the rather
general algebraic
structure established in this paper and the specific representation
of the $W_\infty $ and $W_{1+\infty}$ algebras developed in connection
with two-dimensional quantum gravity as described in in Refs.
\cite{BulloughC1} \cite{BulloughC2}
(In \cite{BulloughC1} \cite{BulloughC2},
these two infinite dimensional algebras
were developed for the ordinary KP hierarchy and included the algebra,
containing the centreless Virasoro algebra, of Ref.\cite{ChengLB}).
In this connection, we note already that if, for example,
$E(R_i)={\rm span}\{A_{im}|\, m\ge 1\},\ i\ge 0,$ and we impose
\[ [A_{im},A_{jn}]=\sum_{l={\rm min}(i-1,j-1)}^{i+j-2}a_l(i-1,j-1,
m-1,n-1)A_{l+1,m+n-1},\]
where the coefficients $a_l$ are defined by
\[ \Bigl [ x^{i+m+1}\frac {d^{i+1}}{dx^{i+1}},
x^{j+n+1}\frac {d^{j+1}}{dx^{j+1}} \Bigl]=
\sum_{l={\rm min}(i,j)}^{i+j}a_l(i,j,m,n)x^{l+m+n+1}\frac {d^{l+1}}{dx^{l+1}}
,\]
then the $E(R)=\sum_{i=0}^\infty E(R_i)$
is a sub-algebra of the $W_{1+\infty}$ algebra of
Refs. \cite{BulloughC1} \cite{BulloughC2} by the identification
$A_{im}=\tau_{i-1,m-1}$; here the $\tau_{i-1,m-1}$ are the elements forming the
$W_{1+\infty}$ algebra \cite{BulloughC1} \cite{BulloughC2} and they may be
realized by $x^{i+m-1}\frac{d^i}{dx^i}$.
\noindent{\bf Acknowledgements:}
One of the authors (WXM)
would like very much to thank the
Alexander von Humboldt Foundation for
the financial support, which
made his visit to UMIST possible.
He is also grateful to Prof. B. Fuchssteiner
for his kind and stimulating discussions.
\small
\begin{thebibliography}{99}
\bibitem{Olver} Olver P J 1980 On the Hamiltonian structure
of evolution equations {\it Math. Proc. Camb. Phil. Soc.} {\bf 88}
71--88
\bibitem{Fuchssteiner1983} Fuchssteiner B 1983
Mastersymmetries, higher order time-dependent symmetries
and conserved densities of nonlinear evolution equations
{\it Prog. Theor. Phys.} {\bf 70} 1508--1522
\bibitem{ChenL} Chen H H and Lee Y C 1985
A new hierarchy of symmetries for the integrable
nonlinear evolution equations
in: {\it Advances in Nonlinear Waves Vol. II} ed. L. Debnath, Research
Notes in Mathematics 111 (Boston: Pitman) p233-239
\bibitem{MaF} Ma W X and Fuchssteiner B 1996
Algebraic structure of
discrete zero curvature equations and
master symmetries of discrete
evolution equations, preprint
\bibitem{ChengLB} Cheng Y, Li Y S and Bullough R K 1988
Integrable non-isospectral flows associated with the Kadomtsev-Petviashvili
equations in $2+1$ dimensions
{\it J. Phys. A: Math. Gen.} {\bf 21} L443-L449
\bibitem{Fushchych} Fushchich (Fushchych) W I, Shtelen W M and Serov N I
1993 {\it Symmetry Analysis and Exact
Solutions of Equations of Nonlinear Mathematical Physics} (Dordrecht: Kluwer)
\bibitem{Fokas} Fokas A S 1987
Symmetries and integrability
{\it Studies in Appl. Math.} {\bf 77} 253--299
\bibitem{Ma1990} Ma W X 1990
$K$-symmetries and $\tau $-symmetries of
evolution equations and their Lie algebras
{\it J. Phys. A: Math. Gen.} {\bf 23}
2707--2716
\bibitem{OevelFZ} Oevel W, Fuchssteiner B and Zhang H W 1989
Mastersymmetries and mutil-Hamiltonian formulations for some integrable
lattice systems
{\it Prog. Theor. Phys.} {\bf 81} 294--308
\bibitem{Ma1992} Ma W X 1992
Lax representations and Lax operator algebras
of isospectral and nonisospectral hierarchies of evolution equations
{\it J. Math. Phys.} {\bf 33} 2464--2476
\bibitem{Tu} Tu G Z 1990
A trace identity and its applications to the theory of discrete
integrable systems {\it J. Phys. A: Math. Gen.} {\bf 23} 3903--3922
\bibitem{Ma1992a} Ma W X 1992
The algebraic structures of isospectral Lax
operators and applications to integrable equations
{\it J. Phys. A: Math. Gen.} {\bf 25}
5329--5343
\bibitem{Kac} Kac V G 1990 {\it Infinite Dimensional Lie Algebras 3rd ed.}
(Cambridge: Cambridge University Press)
\bibitem{OevelF} Oevel W and Fuchssteiner B 1982
Explicit formulas for symmetries
and conservation laws of the Kadomtsev-Petviashvili equations
{\it Phys. Lett. A} {\bf
88} 323--327
\bibitem{ChenLL} Chen H H, Lee Y C and Lin J E 1983
On a new hierarchy of symmetries for the Kadomtsev-Petviashvili equation
{\it Physica D} {\bf 9} 439--445
\bibitem{MaBC} Ma W X, Bullough R K and Caudrey P J 1997
Graded symmetry algebras of time-dependent
evolution equations and application to the modified KP equations,
Dedicated to W. I. Fushchych\footnote
{With sadness we report the death of Professor Wilhelm I.
Fushchych on Monday 7 April 1997 after a short illness. The
remaining three authors of this paper,
Wen-Xiu Ma, Robin Bullough and Philip Caudrey, dedicate this paper
to his memory. Appreciations of W. I. Fushchych will appear in
{\it J. Nonlinear Math. Phys.} of which he was Editor in Chief.}
on his 60th birthday, {\it
J. Nonlinear Math. Phys.} to appear
\bibitem{Caudrey} Caudrey P J 1996 Some thoughts on integrating
non-integrable systems
in: {\it Nonlinear Physics: Theory and
Experiment} eds Alfinito E,
Boiti M, Martina L and Pempinelli F
(Singapore: World Scientific) p60--66
\bibitem{BulloughC1} Bullough R K and Caudrey P J 1995
Solitons and the Korteweg-de Vries equation:
Integrable systems in 1834-1995 {\it Acta Appl. Math.} {\bf 39}
193--228 and references therein
\bibitem{BulloughC2} Bullough R K and Caudrey P J 1996
Symmetries of the classical integrable systems and 2-dimensional quantum
gravity: a `map' {\it J. Nonlinear Math. Phys.} {\bf 3} 245--259
and references therein
\end{thebibliography}
\end{document}