Page 71 - Matrices theory and applications
P. 71

3. Matrices with Real or Complex Entries
                              54
                              and γ ≺ β =(β 1 ,... ,β n−1 ).

                                We apply the lemma to the sequences α = λ, β = a.Since γ ≺ a ,the
                              induction hypothesis tells us that there exists a real symmetric matrix S
                              of size (n − 1) × (n − 1) with diagonal a and spectrum γ. From Corollary
                                                          n
                              3.3.2, there exist a vector y ∈ IR and b ∈ IR such that the matrix
                                                                    T
                                                             S   y
                                                       Σ=
                                                             y   b

                              has spectrum λ.Since s n (a)= s n (λ) = Tr Σ = Tr S + b = s n−1 (a )+ b,we
                              have b = a n . Hence, a is the diagonal of Σ.
                                We prove now Lemma 3.4.1. Let ∆ be the set of sequences δ of n−1real
                              numbers satisfying
                                                                                          (3.4)
                                                α 1 ≤ δ 1 ≤ α 2 ≤ ··· ≤ δ n−1 ≤ α n
                              together with
                                                  k       k

                                                    δ j ≤   β j ,  ∀k ≤ n − 2.            (3.5)
                                                 j=1     j=1
                              We must show that there exists δ ∈ ∆ such that s n−1 (δ)= s n−1 (β ). Since

                                                                               n
                              ∆ is convex and compact (it is closed and bounded in IR ), it is enough to
                              show that

                                              inf s n−1 (δ) ≤ s n−1 (β ) ≤ sup s n−1 (δ).  (3.6)
                                             δ∈∆                    δ∈∆

                              On the one hand, α =(α 1 ,... ,α n−1 ) belongs to ∆ and s n−1 (α ) ≤

                              s n−1 (β ) from the hypothesis, which proves the first inequality in (3.6).
                                Let us now choose a δ that achieves the supremum of s n−1 over ∆. Let r

                              be the largest index less than or equal to n−2 such that s r (δ)= s r (β ), with
                              r = 0 if all the inequalities are strict. From s j (δ) <s j (β )for r< j < n−1,

                              one has δ j = α j+1 , since otherwise, there would exist 	> 0such that
                              ˆ         j                                     ˆ
                              δ := δ + 	e belong to ∆, and one would have s n−1 (δ)= s n−1 (δ)+ 	,
                              contrary to the maximality of δ. Now let us compute

                                    s n−1 (δ) − s n−1 (β )  = s r (β) − s n−1 (β)+ α r+2 + ··· + α n
                                                      = s r (β) − s n−1 (β)+ s n (α) − s r+1 (α)
                                                      ≥ s r (β) − s n−1 (β)+ s n (β) − s r+1 (β)
                                                      = β n − β r+1 ≥ 0.

                              This proves (3.6) and completes the proof of the lemma.
   66   67   68   69   70   71   72   73   74   75   76