50 CHAPTER 2. OVERVIEWOF THE UNDERLYING THEORY
and two block structures, 1(compatible with M11)and
2(compatible with M22).
There are twop erturbed subsystems that wecan study here: Fu(M,1),where ∆1is
closed in a feedback loop around M11;andF
l
(M, 2), where ∆2is closed in a feedback
loop around M22.
We havealready seen that in the case of a dynamic system, the robust stability of
Fu(M,1) is analyzed by checking that µ1(M11)<1. Herewe have used µ1to indicate
that we are considering it with respect to the block structure 1. In the constant
matrix case, we say that the LFT, Fu(M,1) is well posed for all ∆1B∆1if and only
if µ1(M11)<1. This simply means that the inverse in the LFT equations is well defined
for all ∆1B∆1.
The well posedness discussion above applies equally well to Fl(M,2) and we will
denote the µtest for M22 by µ2(M22). However,instead of looking at µ2of M22 ,we
want to look at µ2(Fu(M,1)). Note that Fu(M,1) has the same dimensions as M22
and in fact Fu(M,1)=M
22 when ∆1= 0. In otherwords, what happens when we
apply the µ2test to the whole set of matrices generated by Fu(M,1).
Toanswer this question we need to introduce a larger block structure, denoted here
simply by . This is simply the diagonal combination of the previous twostructures:
= diag(1,2).
Note that this has compatible dimensions with Mitself and the associated µtest will be
denoted by µ(M). Now we can answer the question about what happens to
µ2(Fu(M,1)) for all ∆1B∆1.
Theorem 9 (Main Loop Theorem)
µ(M)<1if and only if
µ1(M11)<1
and
max
1B∆1
µ2[Fu(M,1)] <1
This theorem underlies the fact that robust performance is a simple extension of robust
stability. It has a much more significant role in developing connections between the µ
theory and other theoretical aspects of control. The example in the following section is
an illustration of this point.