Page 72 - Matrices theory and applications
P. 72

3.4.1 Hadamard’s Inequality
                              Proposition 3.4.2 Let H ∈ H n be a positive semidefinite Hermitian
                              matrix. Then
                                                                n

                                                                  h jj .
                                                       det H ≤
                                                               j=1
                              If H ∈ HPD n , the equality holds only if H is diagonal. 3.5. Exercises  55
                                Proof
                                If det H = 0, there is nothing to prove, because the h jj are nonnegative
                                                       j
                              (these are numbers (e ) He ). Otherwise, H is positive definite and one
                                                 j ∗
                              has h jj > 0. We restrict attention to the case with a constant diagonal
                                                   −1/2     −1/2
                              by letting D := diag(h 11  ,... ,h nn ) and writing (det H)/(  j  h jj )=
                              det DHD =det H , where the diagonal entries of H equal one. There



                              remains to prove that det H ≤ 1. However, the eigenvalues µ 1 ,... ,µ n of

                              H are strictly positive, of sum n. Since the logarithm is concave, one has
                                       1            1               1

                                         log det H =     log µ j ≤ log   µ j =log 1 =0,
                                       n            n               n
                                                       j
                              which proves the inequality. Since the concavity is strict, the equality holds
                              only if µ 1 = ··· = µ n = 1, but then H is similar, thus equal to I n .In that

                              case, H is diagonal.
                                Applying proposition 3.4.2 to matrices of the form M M or MM ,one
                                                                                         ∗
                                                                               ∗
                              obtains the following result.
                              Theorem 3.4.3 For M ∈ M n (CC), one has
                                                          1/2
                                                                                     1/2
                                            n   n                         n     n
                                                       2                             2

                                 | det M|≤        |m ij |    ,  | det M|≤      |m ij |  .
                                           i=1  j=1                      j=1  i=1
                              When M ∈ GL n (CC), the first (respectively the second) inequality is an
                              equality only if the rows (respectively the columns) of M are pairwise
                              orthogonal.
                              3.5 Exercises
                                1. Show that the eigenvalues of skew-Hermitian matrices, or as well
                                   those of real skew-symmetric matrices, are pure imaginary.
                                2. Let P, Q ∈ M n (IR) be given. Assume that P + iQ ∈ GL n (CC). Show
                                   that there exist a, b ∈ IR such that aP + bQ ∈ GL n (IR). Deduce that
                                   if M, N ∈ M n (IR) are similar in M n (CC), then these matrices are
                                   similar in M n (IR).
   67   68   69   70   71   72   73   74   75   76   77