
\section{COMPUTER OPERATOR ALGEBRA}
\medskip

%{\bf MAGMA--SWEDES ?? address }

\noindent
\subsection{A Computer Package:\ \ NCAlgebra}
\label{ncalg}
\smallskip
NCAlgebra \cite{HMil2} is a collection of ``functions" for Mathematica 
designed to facilitate manipulation and reduction of noncommutative 
algebraic expressions in operator theory and engineering systems.

We think of this package as a competitor to the yellow pad. Once you 
get used to it this might be considerably more effective for hand
calculations of modest size.  Like Mathematica, the emphasis is on
interaction with the program and flexibility.  

SYSTEMS is a collection of files which run under NCAlgebra 
developed by Mike Walker and Bill Helton.  
This software aims toward the objective of doing 
Hamiltonian (energy balance) calculations
of the type found in $H^\infty$ control for linear 
and certain nonlinear systems.  This effort started with
a paper 
``executable'' under NCAlgebra 
which lead to \cite{BHW2} 
which solves a collection of nonlinear $H^\infty$ 
control problems.  

With regard to extending this work, our inclination is not 
toward breadth. We add to the code in the 
course of doing research by trying to 
produce a collection of broadly 
useful files. Also, the work proposed in the next
section will produce many additions to NCAlgebra.

Others announce that they will produce interactive 
noncommutative packages.
%One is MAGMA by ?? which stands alone, while
%another is NCC by Persson and Sliverstrov at U. of
%Umea, Sweden, which runs under Maple.
Our NCAlgebra is Mathematica's entry in the area and they recommend it
widely. According to Mathematica records about 200 people
have taken it from MathSource.

\subsection{Simplification of Expressions
(with John Wavrik and Mark Stankus)   
}
\label{section:alg}

One objective is to find ``the" rules one would want 
for simplification of
expressions which occur in particular specialties.

The purpose of generating such a list of rules (or a 
simplification theory) is to provide a means for 
replacing complex expressions by expressions which are ``simpler" in 
some sense. The main task is to obtain a list of rules each of which 
replaces a ``complicated" monomial which occurs in an expression by a 
sum of ``simpler" monomials. To simplify such an expression one applies 
the rules to the expression until no further reduction is possible. 

Now we engage in a abbreviated description of our
objectives aimed at an operator theorist. See \cite{HSW}
for less brevity and \cite{HW} for details.
Apologies to Gr\"obner Basis experts for the unconventional
and not wholy consistent presentation. The presentation
would be consistent if we relabeled some expressions
like $x^{-1}$ to be an indeterminant $a$ and
rewrite $x x^{-1} - 1 = 0$ to $x a -1 = 0$
and view all of the computations as sitting
in a noncommutative polynomial ring.


\subsubsection{Replacement Rules}

A replacement rule consists of a left hand side ($LHS$),  which will
always be a monomial, and a right hand side ($RHS$) which will always
be a polynomial. 
Our notation is 
$LHS \rightarrow RHS$.  
Naturally we are unwilling to substitute $RHS$ for  $LHS$ unless
the equation
$LHS-RHS=0$
is true.
Thus replacement rules are associated with equations.
%which in algebraists jargon would be referred to as relations
%on the objects in the equations.

%Conversely if a polynomial with n terms (that is a sum of $n$ monomials) equals
%zero then there are $n$ possible rules we could associate to this
%relation depending on which monomial we use on the $LHS$.

We first give a simple illustration of the reduction of
polynomials by a list of rules.
The definition of 
$(1-xy)^{-1}$ implies that
$(1-xy)(1-xy)^{-1}=1$ which we write in the form
\begin{eqnarray}
\label{eq:1pt1}
xy (1-xy)^{-1}-(1-xy)^{-1} + 1=0 \, ,
\end{eqnarray}
which states that 
a polynomial 
%\footnote{the polynomial is $xyz-z+1$ where we take
%$z=(1-xy)^{-1}$.}
in $x$, $y$ and $(1-xy)^{-1}$ equals 0.
The 3 replacement rules we could associate to this are
\begin{equation}
1\rightarrow(1-xy)^{-1}-xy(1-xy)^{-1} \ \ \ \ 
(1-xy)^{-1}\rightarrow xy(1-xy)^{-1} + 1 
\label{eq:1pt2}
\end{equation}
\begin{equation}
xy (1-xy)^{-1}\rightarrow(1-xy)^{-1} - 1 \, .
\label{eq:1pt4}
\end{equation}
%If we wanted to make expressions more complicated we would choose the first
%(which would make gastly messes) or the second. 
We wish to use the 
rule to make expressions less complicated so we choose the last rule
which replaces the ``most complicated" monomial in (\ref{eq:1pt1}) by a sum of simpler ones.
Now we apply this rule to an expression.
It is an algebraic fact that the polynomial 
$$
(1-xy)^{-2}-2xy(1-xy)^{-2}+x y x y(1-xy)^{-2} - 1 
$$
equals $0$. 
We now claim that two applications of the replacement rule 
(\ref{eq:1pt4}) can be used to establish this
fact. The first application converts this 
(after distributing the products over the sums) to
$$
- (1-xy)^{-2} + 2 (1-xy)^{-1} + x y (1-xy)^{-2} - x y(1-xy)^{-1} - 1
$$
and the second application after cancelling like
terms converts this to zero.

\subsubsection{Complete lists of rules}
\label{subseccomplete}

We now come to one of the more basic points which is 
the concept of `completeness' of a list of rules.
%of computer algebra
We illustrate this with an example.

\begin{example}
\label{example:1pt5}
{\rm {
Consider 
$x$, $x^{-1}$ and $(1-x)^{-1}$ 
and rules based on the definition of inverse
\begin{equation}
\begin{array}{ll} 
x x^{-1}\rightarrow 1 & x(1-x)^{-1}\rightarrow(1-x)^{-1} - 1 \\  
x^{-1} x\rightarrow 1 & (1-x)^{-1}x\rightarrow(1-x)^{-1} - 1  \, .
\end{array}
\label{eq:1pt6}
\end{equation}
There are polynomials in
 $x$, $x^{-1}$ and $(1-x)^{-1}$ 
such as
\begin{equation}
x^{-1}(1-x)^{-1} - (1-x)^{-1} x^{-1}
\label{eq:1pt7}
\end{equation}
which equal 0, but clearly the computer can not decide 
that it is 0 through repeated application of
the rules (\ref{eq:1pt6}).
In this case (and most others) one needs an  enlarged set of rules.
Now consider the rules (\ref{eq:1pt6}) together with
the following two rules which are algebraic consequences
of (\ref{eq:1pt6})
\begin{equation}
\begin{array}{ll}
x^{-1} (1-x)^{-1} & \rightarrow x^{-1} + (1-x)^{-1} \\
(1-x)^{-1} x^{-1} & \rightarrow x^{-1} + (1-x)^{-1}
\end{array}
\label{eq:1pt8}
\end{equation}
(and which an operator theorist calls the resolvent identity).
It is clear that (\ref{eq:1pt7}) reduces to 0 by applying
the full list of rules (\ref{eq:1pt6}) and (\ref{eq:1pt8})
repeatedly (actually one pass will do).
% and you 
%do not really need (\ref{eq:1pt6})).
}}
\end{example}

More generally one can prove 
that when this full list of rules is applied repeatedly
to any polynomial $p$
in $x$, $x^{-1}$ and $(1-x)^{-1}$ 
one obtains one particular reduced polynomial $q$  {\it regardless of the 
order in which the rules are applied}. That is $q$ is a 
``canonical form" 
for $p$, which is in a certain sense as simple as possible.
Also, one can prove that {\it if it is possible to derive  using only algebra and the definitions of
 $x^{-1}$ and $(1-x)^{-1}$ that $p=0$, then
it's canonical form
is zero}.

Thus the set of rules (\ref{eq:1pt6}) and (\ref{eq:1pt8}) are ``complete"
in a profound sense, we henceforth denote these by (RESOL).
Such a set of rules is called a {\it Gr\"obner basis} or {\it GB} for 
$x$,$x^{-1}$,$(1-x)^{-1}$
with a particular ordering on monomials.
There is not space to discuss orderings. Intuitively, one places
an ordering on monomials with complicated monomials high in the
order. In specialized terms, we use graded lexicographic
order induced by $x < x^{-1} < (1-x)^{-1}$.
We shall not give a formal definition of a noncommutative GB 
but refer the reader to the works of F.Mora, 
%T.Mora 
or {\bf [HW]}.

It is interesting to note that we teach beginning
students in functional analysis the resolvent
identity which is essentially (RESOL).
Thus, by tradition, operator theorists 
inadvertently discovered the GB (RESOL).

\vskip 0.2 in

\subsection{Further Common Situations}

We have also found GB for
reducing polynomials in
\footnote{Here we use graded lexicographic order consistent
with the order in which the symbols are listed.}
\begin{description}
\item{(EB)} \ \
$x$, $y$,  $x^{-1}$, $y^{-1}$, $(1-xy)^{-1}$ and $(1-yx)^{-1}$

\item{(preNF)} \ \ 
$x$, $y$,  $x^{-1}$, $y^{-1}$, $(1-x)^{-1}$, $(1-y)^{-1}$,
$(1-xy)^{-1}$ and $(1-yx)^{-1}$.

\item{(NF)} \ \ 
$x$, $y$,  $x^{-1}$, $y^{-1}$, $(1-x)^{-1}$, $(1-y)^{-1}$,
$(1-xy)^{-1}$, $(1-yx)^{-1}$, $(1-xy)^{1/2}$ and $(1-yx)^{1/2}$.
\end{description}

These objects were chosen for extensive study because they
form the cornerstone of computations in various subjects.
The expressions (EB) occur in energy balance formulas in $H^{\infty}$ control.
See section \ref{section:four}. These combined with basic inverses often called resolvents
constitute (preNF). 
The relations (NF) are
extremely important to those working with 
$2\times2$ block unitary matrices or with the 
Nagy-Foias operator model
which is the paradigm for discrete time energy conserving systems.
%John Wavrik and I analyzed these problems by running his
%program which implements a noncommutative Buchberger algorithm,
%Then came the difficult part of seeing what the pattern was.

We found that there is a marked change of structure in each problem. 

\begin{result}
\label{EBbasisFinite}
(EB) gives finite Gro\"bner basis
while (preNF) and (NF) do NOT. 
\end{result}

\begin{result}
\label{preNFbasisInfinite}
The GB for (preNF) is infinite.
However, this GB can be parameterized
by a finite list of formulas. (There are 8 parameterization
laws plus a handful of special cases.) The (GENR) rules above contain 
about half of these formulas.
The GB for (NF) has a similar structure.
\end{result}

Result \ref{EBbasisFinite} is proven and 
Result \ref{preNFbasisInfinite} was
conjectured in {\bf {[HW]}} and
work in progress with Mark Stankus proves 
Result \ref{preNFbasisInfinite}. 


\vskip 0.2in

\subsection{General Rules}
In operator theory
in addition to resolvents one
often studies operators of the form
$h(xy)$ where $h$ is a function analytic on the spectrum of $xy$ and $yx$.
This is called the {\it 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.

We teach our students the extremely valuable rule
$$h(xy)x \rightarrow xh(yx) \, ,$$
so it is amusing to ask whether we 
teach our students
enough rules. Wavrik, Stankus and I maintain students should also
be taught the list of rules 

\smallskip
{\centerline {GENR Rules}}
$$
h(xy) x\rightarrow x h(yx) \quad
h(yx) x^{-1}\rightarrow x^{-1}  h(xy)
\leqno{(Gr1)}
$$
$$
x h(yx) (\lambda-x)^{-1}\rightarrow \lambda h(xy) (\lambda-x)^{-1} - h(xy)
\leqno{(Gr2)}
$$
%Cosmetic change to Gr3...MXS
$$
x^{-1} h(xy) (\lambda-x)^{-1}\rightarrow \lambda^{-1} x^{-1} h(xy)
+\lambda^{-1} h(yx) (\lambda-x)^{-1}
\leqno{(Gr3)}
$$
%Corrected power of lambda in Gr4...MXS
$$
(\lambda-x)^{-1} h(y x) (\lambda-x)^{-1}-(\lambda-x)^{-1} h(x y) 
(\lambda-x)^{-1} \rightarrow 
\leqno{(Gr4)}
$$
$$
\lambda^{-1} (h(y x) (\lambda-x)^{-1}-(\lambda-x)^{-1} h(x y) ) \, .
$$
$$
\mbox{\rm The\ rules\ (Gr1)\ through\ (Gr4)\ with\ $x$ and $y$ swapped.}
\leqno{(Gr5)}
$$

The (GENR) equations are easy to verify, but the hard question is
whether or not this is a complete set.
Our analysis of this requires a long explanation the flavor of which is:

\begin{result}
\label{GENRResult}
We associate to this an
algebraic setting  and a natural (partial) order.
Then we find a list of rules (which is infinite) and
prove by hand that it is a GB. This has the interpretation
that (GENR) is a `complete' list of rules which are true for all
$h$ and $x$,$y$. 
\end{result}

The result is discussed much more thoroughly in {\bf [HW]}. 
(GENR) is not a GB when one specifies a
particular $h$. 
For example, if $h(xy) = (1-xy)^{-1}$, the replacement 
rule $xyh(xy) \rightarrow h(xy) - 1$ is not a consequence of the rules in the 
expanded set even though it holds for this particular $h$. On the 
other hand, this rule is not universally applicable;
it fails for many analytic functions $h$. One should think of 
(GENR) as a complete set of universally applicable rules. 
We have found that extremely effective simplification can be 
performed just using the expanded set supplemented by some of the 
obvious rules for a particular $h$. 

Currently we are investigating something called
a Universal Gr\"obner basis for (GENR). This is a 
larger GB than (GENR) and is not sensitive to
the graded lexicographic order used on monomials.
We have a short list of formulas and we are trying
to prove that it is a Universal GB.

Also we are studying  simplification of the formulas
occurring in continuous time lossless systems paradigms 
(Brodski-Livsic systems), in much of $H^\infty$ control and in other areas. 

\subsubsection{Methods}
\label{section:four}
 
   In the commutative setting, there is the algorithm of
Buchberger which always terminates and provides a Gr\"obner Basis.
We have used Mora's algorithm which is an adaptation 
of the Buchberger algorithm to polynomials in non-commuting 
variables due to F. Mora in 1986. 

The details of this algorithm are found in {\bf [HW]}. 
In contrast to the 
commutative case, the Mora Algorithm does not always terminate.
In this case, Mora's algorithm does not guarantee
a GB. Indeed what one sees on the computer
is that the disk fills with a collection of rules which
to a  system theorist look meaningless. They, of course,
might be very practical, but then again maybe not.
In the early stages of the subject one must analyze a
few of cases thoroughly to see if the structure 
is perverse or orderly. Our experience indicates it is orderly.
Most of our early simplification work used Wavrik's GB code (written in forth).
 
\subsubsection{Discovering Theorems Semi-automatically}
\label{subsubsection:theoremdiscovery}

In another direction we introduce a notion
of {\it strategy}. 
A strategy is an iterative process capable 
of deriving some theorems by running Mora's GB algorithm,
prompting for a small isolated piece of human
intervention, then running the algorithm again and repeating. 
The (somewhat vague) question currently being tested is which
theorems from the core above can be derived using
a `1-strategy' or a `2-strategy'.
The 1 or 2 etc. measures the human intervention in
a simple way.
The objective here is to develop
not just a systematic collection of new proofs
of old theorems, but a tool for helping
a researcher discover new theorems in
system theory. This is very ambitious
and even if partly successful should be very useful for
some classes of problems.
At least conversion from input-output to statespace coordinates,
often a considerable challenge, might 
become easier. One test case produces 

\begin{result} 
\label{BGKF}
The classic theorem 
of Bart-Gohberg-Kaashoek-VanDoren 
setting
factorization of systems in correspondence with
projections which intertwine certain 
operators (invariant subspaces) can be
obtained from a 2-strategy.
\end{result}

This first test case envolved considerable
programming and hand calculation.
We are steadily making implementation more automatic and
beginning  many other test cases involving 
factorization of systems. Mathematica (Mma) is slow
but we run a GB algorithim repeatedly, so 
we will soon have fast C++ code implementing 
a noncommutative GB algorithm.
This interfaces with Mma so user friendlyness
remains and those things which
change in the course of an experiment 
we change in Mma not C++.
The C++ code could link into Maple, but would have 
extremely narrow uses, in the absence of something like NCAlgebra.

In terms of computer algebra what we are doing
in this subsection is studying
not graded but elimination orderings.
In the commutative case these put systems of
polynomial equations in triangular form.
%Then one gets a nested sequence of problems
%$p_j(x_j)=0$true?? in one variable which one
%solves with a root finder.
The noncommuting case is not nice in general.
What we are seeking 
is a natural practical noncommutative generalization.
We were surprised to find that this might
be something which heavily assists one in finding
systems engineering theorems.

All of the operator-system theoretic studies we are pursueing
will have a direct impact on NCAlgebra. In addition to raw 
GB generation we shall permanantly insert a set of 
workhorse rules like those
derived in section \ref{section:alg}. On most simplification
problems these will run first. Then the raw GB algorithm 
will try to find anything missed because of the special
nature of the problem.

\subsubsection{An Application}

Here is a demo which anyone can follow at a high level 
though it is set in the context of $H^\infty$ control which
most readers will have to downplay.  

 In $H^\infty$ control one deals with a Hamiltonian $H$ on the
state space $(x,z)$ of the closed loop system.
Here $x$ is a state of the plant and $z$ is a
state of the compensator. The unknowns in $H$ are a quadratic form
$\varepsilon$ which is to be a storage function 
of the closed loop system and the $a,b,c,d$ of 
the unknown compensator. As usual take $d=0$.
If a solution $\varepsilon$ exists 
one can derive
that some controller, called the
central controller, $a,b,c$ 
must be given by certain formulas. 
These formulas do not imply that a solution to the $H^{\infty}$
control problem exist and to see if it does one must plug the
central controller formulas back into $H$ and see if $H\le 0$
for all states $(x,z)$
of the closed loop system. Let's see how this computation 
goes with the methods we have discussed.

{\tiny{
\begin{verbatim}

H= tp[x] ** X ** A ** z + tp[x] ** inv[Y] ** A ** x - tp[x] ** inv[Y] ** A ** z + tp[x] ** tp[A] ** X ** z + tp[x] ** tp[A] ** inv[Y] ** x - 
  tp[x] ** tp[A] ** inv[Y] ** z + tp[x] ** tp[C1] ** C1 ** x + tp[z] ** X ** A ** x - tp[z] ** X ** A ** z - tp[z] ** inv[Y] ** A ** x + 
  tp[z] ** inv[Y] ** A ** z + tp[z] ** tp[A] ** X ** x - tp[z] ** tp[A] ** X ** z - tp[z] ** tp[A] ** inv[Y] ** x + tp[z] ** tp[A] ** inv[Y] ** z + 
  tp[x] ** X ** B1 ** tp[B1] ** X ** z - tp[x] ** X ** B2 ** tp[B2] ** X ** z + tp[x] ** inv[Y] ** B1 ** tp[B1] ** inv[Y] ** x - tp[x] ** inv[Y] ** 
  B1 ** tp[B1] ** inv[Y] ** z + tp[z] ** X ** B1 ** tp[B1] ** X ** x - tp[z] ** X ** B1 ** tp[B1] ** X ** z - tp[z] ** X ** B2 ** tp[B2] ** X ** x + 
  tp[z] ** X ** B2 ** tp[B2] ** X ** z - tp[z] ** inv[Y] ** B1 ** tp[B1] ** inv[Y] ** x + tp[z] ** inv[Y] ** B1 ** tp[B1] ** inv[Y] ** z - tp[x] ** 
  X ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** x + tp[x] ** X ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** z + tp[x] ** inv[Y] ** 
  inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** x - tp[x] ** inv[Y] ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** z - tp[x] ** tp[C2] ** C2 ** Y ** 
  inv[-1 + X ** Y] ** X ** x + tp[x] ** tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] ** X ** z + tp[x] ** tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] ** 
  inv[Y] ** x - tp[x] ** tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] ** inv[Y] ** z + tp[z] ** X ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** x - tp[z] ** 
  X ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** z - tp[z] ** inv[Y] ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** x + tp[z] ** inv[Y] ** 
  inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** z + tp[z] ** tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] ** X ** x - tp[z] ** tp[C2] ** C2 ** Y ** 
  inv[-1 + X ** Y] ** X ** z - tp[z] ** tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] ** inv[Y] ** x + tp[z] ** tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] **
  inv[Y] ** z + tp[x] ** X ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] ** X ** x - tp[x] ** X ** inv[-1 + Y ** X] ** Y **
  tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] ** X ** z - tp[x] ** X ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] ** inv[Y] ** x + 
  tp[x] ** X ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] ** inv[Y] ** z - tp[x] ** inv[Y] ** inv[-1 + Y ** X] ** Y ** tp[C2] ** 
  C2 ** Y ** inv[-1 + X ** Y] ** X ** x + tp[x] ** inv[Y] ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] ** X ** z + 
  tp[x] ** inv[Y] ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] ** inv[Y] ** x - tp[x] ** inv[Y] ** inv[-1 + Y ** X] ** Y ** 
  tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] ** inv[Y] ** z - tp[z] ** X ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] ** X ** x + 
  tp[z] ** X ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] ** X ** z + tp[z] ** X ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** 
  Y ** inv[-1 + X ** Y] ** inv[Y] ** x - tp[z] ** X ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] ** inv[Y] ** z + tp[z] **
  inv[Y] ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] ** X ** x - tp[z] ** inv[Y] ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** 
  Y ** inv[-1 + X ** Y] ** X ** z - tp[z] ** inv[Y] ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] ** inv[Y] ** x + tp[z] **
  inv[Y] ** inv[-1 + Y ** X] ** Y ** tp[C2] ** C2 ** Y ** inv[-1 + X ** Y] ** inv[Y] ** z
\end{verbatim}
}}

%\baselineskip=18pt 

Here,
with the usual normalization,
$
%\begin{equation}
\varepsilon(x,z)= 
\pmatrix{ x^T z^T}
\pmatrix{
Y^{-1} & -(Y^{-1}-X) \cr
-(Y^{-1}-X) & Y^{-1}-X
}
\pmatrix{x \cr z}
%\label{eq:4.1}
%\end{equation}
$
and we have taken relations called the 
Doyle Glover Kargonekar Francis (DGKF) simplifying 
assumptions which greatly 
reduce the complexity of the formulas.

Here we have used the same notation that 
one finds in our NCAlgebra program to give a feel for 
how this type of computer 
algebra goes.
{\it tp} stands for transpose while {\it inv}
stands for inverse and `**' for multiply.
%and {\it XX}, {\it YY}, {\it YYI} stand for X, Y, YYI
%respectively. 
%{\bf change to X? and Y?? YES DO IT}

The rules (RESOL) together with those derived for (EB) are stored in a 
function NCSimplifyRational in NCAlgebra which applies them 
repeatedly to an expression until no change occurs. We get the considerably simpler expression.

{\tiny{
\begin{verbatim}
Hsimple = NCSimplifyRational[H]=
  tp[x] ** X ** A ** z + tp[x] ** inv[Y] ** A ** x - tp[x] ** inv[Y] ** A ** z + tp[x] ** tp[A] ** X ** z + tp[x] ** tp[A] ** inv[Y] ** x - 
   tp[x] ** tp[A] ** inv[Y] ** z + tp[x] ** tp[C1] ** C1 ** x - tp[x] ** tp[C2] ** C2 ** x + tp[x] ** tp[C2] ** C2 ** z + 
   tp[z] ** X ** A ** x - tp[z] ** X ** A ** z - tp[z] ** inv[Y] ** A ** x + tp[z] ** inv[Y] ** A ** z + tp[z] ** tp[A] ** X ** x - 
   tp[z] ** tp[A] ** X ** z - tp[z] ** tp[A] ** inv[Y] ** x + tp[z] ** tp[A] ** inv[Y] ** z + tp[z] ** tp[C2] ** C2 ** x - 
   tp[z] ** tp[C2] ** C2 ** z + tp[x] ** X ** B1 ** tp[B1] ** X ** z - tp[x] ** X ** B2 ** tp[B2] ** X ** z + 
   tp[x] ** inv[Y] ** B1 ** tp[B1] ** inv[Y] ** x - tp[x] ** inv[Y] ** B1 ** tp[B1] ** inv[Y] ** z + tp[z] ** X ** B1 ** tp[B1] ** X ** x - 
   tp[z] ** X ** B1 ** tp[B1] ** X ** z - tp[z] ** X ** B2 ** tp[B2] ** X ** x + tp[z] ** X ** B2 ** tp[B2] ** X ** z - 
   tp[z] ** inv[Y] ** B1 ** tp[B1] ** inv[Y] ** x + tp[z] ** inv[Y] ** B1 ** tp[B1] ** inv[Y] ** z 
\end{verbatim}
}
}

Notice that everything of the form inv[1-Y**X] and
inv[1-X**Y] have been eliminated from $H$.
In general we have found NCSimplifyRational 
very effective in simplifying expressions with 
inverses of (very simple) polynomials.

This is still a bad expression, but we have not yet used 
the fact that
$X$ and $Y$ solve the two famous DGKF Riccati equations
$
X A + A^T X + C_1^T C_1 + 
X  B_1 B_1^T X - X B_2 B_2^T X = 0
 ,  \ \ 
Y^{-1} A + A^T Y^{-1} + C_1^T C_1 - 
C_2^T C_2 + Y^{-1} B_1 B_1^T Y^{-1} = 0 \,  .
$

Now we show the effect of our simplification methods.
First set an ordering on the indeterminants in these
expressions.
Say 
$ A <  B_1 < B_2 < C_1 < C_2 < x < X < Y^{-1} < 
z < A^T < B_1^T < B_2^T < C_1^T < C_2^T < x^T < z^T$.

Now we invoke the NCAlgebra command GroebnerSimplify.
This command finds a GB for the 
DGKF Riccati equations and reduces $Hsimple$
by that Gr\"obner basis. The command 
reduced $Hsimple$ to 0. 

This completes the core of the sufficient 
side of the famous DGKF theorem. Namely, that the central
controller has desired $H^{\infty}$ performance level.
It was done automatically with generic methods.

Note in this example we computed a finite 
GB (it contains four relations) for  a collection of symbols (matrices)
satisfying the two DGKF Riccati equations and for a particular order.
The order was chosen by default in the middle
of a computer session. Since it worked, we did not explore
other orders.
%Incidentally, the GB contains
%four relations.


