Another way of combining updates is by chosing an update mechanism at random
and then implementing that update, thus producing a statistical mixture of
the updates. If update steps described by transition kernels 
are chosen
with probabilities 
then the transition kernel of the mixed update step
is 
. The sum can be finite or infinite.
In order for the to specify a probability distribution, the must
be nonnegative and sum to one. Thus mixing updates corresponds to the
operation of forming a convex combination of kernels. Again, it is obvious
that if 
is stationary for each of the , then is also
stationary for 
.