\documentclass[12pt,thmsa]{article}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{document}

\section{GENERAL RULES}

We find that relations of the form $h(xy)x=xh(yx)$ and $yh(xy)=h(yx)y$ hold
for many analytic functions $h$. We have seen this in the preNF rules where $%
h(z)=(1-z)^{-1}$and we note that these relations also hold when $%
h(z)=(1-z)^{-n}$. We therefore applied the machinery in this paper to study
this general situation. The result is a set of simple and general
substitution rules. These include some of the rules obtained in the previous
sections. This set of rules holds for all operators $x$ and $y$ on a Hilbert
space $\mathcal{H}$ with $x$, $y$, $\lambda -x$ and $\lambda -y$ invertible,
functions $h$ analytic on the spectrum of $xy$ and $yx$ and all $\lambda \ne
0$. (Technically, the following are not necessarily replacement rules for
certain $h$ since the left hand side may not be a monomial. We will discuss
this shortly.)

\bigskip

\subsection{Statement of the (GENR) Rules}

\centerline{GENR Rules} 
$$
h(xy)x\rightarrow xh(yx)\leqno{(Gr0)} 
$$
$$
h(yx)x^{-1}\rightarrow x^{-1}h(xy)\leqno{(Gr1)} 
$$
$$
xh(yx)(\lambda -x)^{-1}\rightarrow \lambda h(xy)(\lambda -x)^{-1}-h(xy)%
\leqno{(Gr2)} 
$$

$$
x^{-1}h(xy)(\lambda -x)^{-1}\rightarrow \lambda ^{-1}x^{-1}h(xy)+\lambda
^{-1}h(yx)(\lambda -x)^{-1}\leqno{(Gr3)}
$$
$$
(\lambda -x)^{-1}h(yx)(\lambda -x)^{-1}-(\lambda -x)^{-1}h(xy)(\lambda
-x)^{-1}\rightarrow \leqno{(Gr4)} 
$$
$$
\lambda ^{-1}(h(yx)(\lambda -x)^{-1}-(\lambda -x)^{-1}h(xy)) 
$$
$$
\mbox{\rm The\ rules\ (Gr0)\ through\ (Gr4)\ with\ $x$ and $y$ swapped.}%
\leqno{(Gr5-9)} 
$$

Operators of the form $h(xy)$ arise, of course, in what is called the 
\textit{functional calculus} of the operator $xy$. Examples are $%
h(xy)=(1-xy)^{-1}$, $h(xy)=(1-xy)^{1/2}$ and $h(xy)=e^{xy}$ provided the
eigenvalues of $xy$ are in the right location.

In the above list of rules, the expression on the left hand side of the rule
may not be a monomial. For example, see (Gr4) or set $h(z)=z+z^2$. It is
easier to automate replacement rules if the $LHS$ of a rule is a monomial.
When these rules are used for machine computation, they are rearranged,
using a term ordering, so that $LHS$ is a monomial.

This set of rules applies to many situations. Currently (RESOL) is
universally taught; (RESOL4) is called the resolvent identity. We also teach
students the exremely valuable rule 
$$
h(xy)x\rightarrow xh(yx)\,,
$$
but (GENR) is not taught. (GENR) is a short and powerful enough list of
rules that it could be used effectively in simplification by hand.

\subsection{Genesis of the (GENR) Rules}

We introduce variables $x<y<x^{-1}<y^{-1}<(1-x)^{-1}<(1-y)^{-1}<S<T$ and use
the defining relations for the inverses together with the relations $Sx=xT$
and $yS=Ty$. An infinite Basis was obtained and it was shown to be a
Gr\"{o}bner Basis (see [HW] for details). This Basis has the property,
however, the infinite families of rules involving $S^n$ and $T^n$ are all
obtained by substituting $S^n$ for $S$, and $T^n$ for $T$ in a finite subset
of the rules. This subset we have given as (GENR). We have replaced $S$ by $%
h(xy)$ and $T$ by $h(yx)$. Since the starting relations hold in this
setting, the (GENR) rules hold under the conditions above. Since $h(xy)^n$
and $h(yx)^n$ also satisfy the starting conditions, all of the rules in the
infinite family are subsumed. The (RESOL) rules together with (GENR) rules,
if applied in this way to powers, give a set of rules which is complete in
the sense that all algebraic consequences of our starting relations reduce
to $0$.

It should be noted that it is also easy to verify the (GENR) rules by hand,
so that they can be introduced independent of the Gr\"{o}bner Basis
machinery. For example, the following calculation verifies (Gr3).

\textbf{?? What are you taking as hypotheses here? Doesn't the first step
use Gr1 and the second RESOL?} 
\begin{eqnarray*}
x^{-1}h(xy)(\lambda -x)^{-1} &=&h(yx)x^{-1}(\lambda -x)^{-1} \\
&=&h(yx)(\lambda ^{-1}x^{-1}\ +\ \lambda ^{-1}(\lambda -x)^{-1}) \\
&=&\lambda ^{-1}h(yx)x^{-1}\ +\ \lambda ^{-1}h(yx)(\lambda -x)^{-1}
\end{eqnarray*}

\subsection{Properties of GENR}

The GENR rules are interesting in that they show that the technology we have
developed can produce results of considerable generality. While the
algebraic part of this technology just involves polynomials, substitution
for the variables can be made in a variety of settings.

To be sure, however, the starting relations only embody the simple
assumptions that, whatever $h(xy)$ and $h(yx)$ happen to be, we have 
\begin{eqnarray*}
(*)\;h(xy)x &=&xh(yx) \\
yh(xy) &=&h(yx)y
\end{eqnarray*}
$.$ The GENR rules follow as an algebraic consequence of these assumptions.
The rules are, therefore, valid for any $h(xy)$ and $h(yx)$ for which these
conditions hold.

Now suppose we specialize to a particular $h$. If, for example, $%
h(z)=(1-z)^{-1}$we will have the same atomic expressions as we used in the
preNF situation. A major difference, however, is that in the preNF case we
added additional relations which come from the definition of $(1-xy)^{-1}$
and $(1-yx)^{-1}$. In fact, the conditions (*) can be derived algebraically
from these added conditions (they appear in the GB for preNF). We do not
expect, nor do we find, that GENR embodies all the extra relations that may
hold for particular $h$. What we do expect is that GENR will provide a
useful set of easily implemented rules that at least provides simplification
in a general sense -- without using any special properties of a particular $%
h $.

\subsection{Comparison of GENR with preNF}

Of the 8 infinite families listed for preNF, we find that I, IV, V, and VIII
can be obtained from the GENR rules, while the other 4 families cannot. Thus
the first listed families must be consequences of just the relations (*) and
the defining relations for the inverses. The remaining 4 families require
the use of more particular properties of $(1-xy)^{-1}$ and $(1-yx)^{-1}$

\textbf{?? If you sort out what happens for the special preNF relations,
that information can be included here.. }

\end{document}
