58 Noncommutative Mathematics for Quantum Systems
then X+Y is essentially self-adjoint and the distribution w.r.t.Ωof its
closure is equal to the additive free convolution of the distributions of X
and Y w.r.t. toΩ, i.e.
L(X+Y,Ω) =L(X,Ω)L(Y,Ω).
One can show that the reciprocal Cauchy transform Fμ of a
probability measureμ ∈ M 1 (R)is invertible in an appropriate
domain, and define the so-calledVoiculescu transformofμas
φμ(z) =F〈−^1 〉(z)−z,
whereF〈−^1 〉denotes the inverse ofF. This transform linearizes the
free convolution, we have
φμν=φμ+φν.
as can be checked using the ‘parallelogram equation’ in Theorem
1.7.3.
Example 1.7.5 Consider the Bernoulli distribution
μ=
1
2
(δ+ 1 +δ− 1 ).
Then we compute
Gμ(z) =
z
z^2 − 1
and Fμ(z) =z−
1
z
.
Therefore,
Fμ〈−^1 〉(z) =
1
2
(
z+
√
z^2 + 4
)
and
φμμ(z) = 2 φμ(z) = 2
(
Fμ〈−^1 〉(z)−z)
)
=
√
z^2 + 4 −z
forzwith Im(z)>1. We get
Fμμ(z) =
√
z^2 − 4
and see that the free convolution of two Bernoulli distributions is
an arcsine distribution,
μμ=
1
π
1
√
4 −x^2
(^1) ]−2,2[(x)dx.