Page 87 - Matrices theory and applications
P. 87

4. Norms
                              70
                              4.4 A Lemma about Banach Algebras
                              Definition 4.4.1 A normed algebra is a K-algebra endowed with a norm
                              satisfying  xy ≤  x  y . Such a norm is called an algebra norm.When
                              a normed algebra is complete (which is always true in finite dimension), it
                              is called a Banach algebra.
                              Lemma 4.4.1 Let A be a normed algebra and let x ∈A. The sequence
                                      m 1/m
                              u m :=  x      converges to its infimum, denoted by r(x). Additionally,
                              if K = CC,andif A has a unit element and is complete, then 1/r(x) is
                              the radius of the largest open ball B(0; R) such that e − zx is invertible for
                              every z ∈ B(0; R).
                                Of course, one may apply the lemma to A = M n (CC) endowed with
                              a matrix norm. One then has r(x)= ρ(x), because e − zx = I − zA is
                              invertible, provided that z is not the inverse of an eigenvalue. In the case
                              K = IR, one uses an auxiliary norm N that is the restriction to M n (IR)of
                              an induced norm on M n (CC). Since  ·   and N are equivalent, one simply
                              writes
                                                                              m 1/m
                                                             m 1/m
                                                  m 1/m
                                        ρ(A)= ρ(A )     ≤ A        ≤ C 1/m N(A )   .
                              The latter sequence converges to ρ(A) from the lemma, which implies the
                              convergence of the former. We thus have the following result.
                              Proposition 4.4.1 If A ∈ M n(K),then
                                                                  m 1/m
                                                    ρ(A) = lim  A
                                                           m→∞
                              for every matrix norm.
                                Proof

                              Convergence. The result is trivial if x m  = 0 for some exponent. In the
                                   opposite case, we use the following inequalities, which come directly
                                   from the definition:
                                                            p a
                                                                 r
                                                 x ap+r  ≤  x    x  ,  ∀a, p, r ∈ IN.
                                   We then define
                                                          1
                                                                 m
                                                     v m =  log  x   =log u m .
                                                          m
                                   Let us fix an integer p and perform Euclidean division of m by p:
                                   m = ap + r with 0 ≤ r ≤ p − 1. This yields
                                                               apv p + rv r
                                                        v ap+r ≤         .
                                                                 ap + r
                                   As m, hence a, tends to infinity, the right-hand side converges,
                                   because rv r remains bounded:

                                                          lim sup v m ≤ v p .
   82   83   84   85   86   87   88   89   90   91   92