Page 107 - Special Topic Session (STS) - Volume 3
P. 107

STS518 Steen T.
            chosen  vector  state).  If   is  unbounded  (i.e.  non-continuous)  the  spectral
            distribution µ  is similarly defined by the equation
                         
                                       [()] = ∫ ()  ()
                                                         
                                       
                                                  ℝ
            which is then required to hold for any bounded Borel function  ∶  ℝ  →  ℝ,
            and where () is defined in terms of spectral calculus (see e.g. [17]). In this
            case µ  generally has unbounded support.
                  
                Each of the five notions of independence mentioned above gives rise to a
            corresponding notion of (additive) convolution of two probability measures µ
            and  on ℝ, roughly defined as the (spectral) distribution of   +  , where 
            and  are two Hermitian operators such that µ  =  µ, µ  =  , and  and  are
                                                         
                                                                 
            independent in the considered sense (see Theorem 2.1 below).
                Among  the  four  non-classical  notions  of  independence,  free
            independence,  introduced  by  Voiculescu  in  the  1980’s,  is  by  far  the  most
            developed and well-studied. This is mainly due to its applications in the theory
            of  operator  algebras  (which  was  Voiculescu’s  original  motivation  for
            introducing the concept; see [22]) and its striking connection to the theory of
            random  matrices,  which  we  shall  indicate  below.  In  the  Hilbert  space
            framework  outlined  above,  two  Hermitian  bounded  operators  and  are
            freely  independent,  if,  for  any  sequence  ,  ,  , . .. of  polynomials  in  one
                                                       1
                                                          2
                                                             3
            variable and for any  in , it holds that

            and that


            and similar conditions for products starting with polynomial expressions in y.
            In  words  the  requirement  is  that  any  product  of  centered  polynomial
            expressions, alternating in x and y, must have expectation equal to 0.

            2.  Free additive convolution
            2.1  Theorem  &  Definition  ([21],[7]).  Let µ and ν be  (Borel-)  probability
            measures on ℝ. Then there exists a Hilbert space ℋ, a unit vector ξ in ℋ and
            (possibly unbounded) Hermitian operators x and y acting on ℋ, such that the

            following conditions hold (with the notation introduced above)
              (i)  µ  =  µ, µ  =  ν.
                    x
                            y
              (ii)  x and y are freely independent with respect to E .
                                                                 ξ
                The conditions (i) and (ii) uniquely determine (in particular) the spectral
            distribution of the Hermitian operator x  +  y, which may thus be denoted by
            µ  ν and referred to as the free (additive) convolution of µ and ν.





                                                                96 | I S I   W S C   2 0 1 9
   102   103   104   105   106   107   108   109   110   111   112