Suppose
that the SISO plant
has a linear fractional
dependence on
, namely there are
rational functions a, b, c, d
meeting the
nondegeneracy condition
so that
Suppose the uncertainty
.
Then
and nondegeneracy ensures the two sets do not
intersect.
Let
and
.
The Agler inequalities (3)
have as unknowns two
positive semidefinite
valued
functions
and
on
go as follows.
The pole
interpolation constraints give
The zero interpolation constraints give
The cross term between the two is
We must find
and
meeting these conditions
so that
are positive semidefinite in the following sense
for all
. Here
Likewise for
. Obviously,
a necessary condition which is practical to
check is select any finite subset of
which is not too large, plug it
into the formulas above and thereby get a finite
dimensional LMI.
The Agler inequalities (once we
divide) by the
term have coefficients which are two
valued functions on
.
One coefficient is
which is completely determined
by the performance specs
and
.
The other is
which is completely determined by the plant model. Then the Agler inequalities are
restricted
to
, where
denotes the Schur product (entry by entry
product) of two matrices and
must be
positive semidefinite.
It can be proved using a normal families argument,
that the function
may be taken to be
antianalytic in
and analytic in
.
That is if a nonnegative solution
to
(13)
exists, then one with these analyticity properties exists.
It might be helpful to
note how the classical control problem with no
uncertainty fits in formally.
This is
the case where
.
In this case the
terms involving
drop out, so what
remains is
.
Since
the set
is finite,
namely, the
plant's pole zero set,
By inspection one sees that
restricted to
is a Pick
matrix; indeed we have the old fashion
control Nevanlinna-Pick solution.
The
inequalities (13) suggest an
interpretation
when
.
If
is a solution,
then
Each choice of
gives
with a finite zero
pole set
and the
statement
on
is equivalent to the
statement that an
controller meeting
the specs exists.
Now (15) implies
this and even much stronger conditions like
on
,
conditions so strong as to be likely unmeetable.
Fortunately, (13 )
is more lenient than
(15)
because
has the effect of
lessening the
constraint.