The scaled Dirichlet distribution is the probability distribution obtained by rescaling the coordinates of a Dirichlet
distribution and then renormalizing their sum (Ng et al. 2011). For , parameters
, and
with Dirichlet distribution
of parameters
, set
|
(1)
|
Writing
and
, its probability
density function on
,
, with respect to
, is
|
(2)
|
Here
is the gamma function. Equal
give the ordinary Dirichlet
distribution. Multiplying every
by the same positive constant
leaves the distribution unchanged. The parameters
and the ratios of the
are identifiable, giving
free parameters (Lakhani 2026).
For nonnegative integers , put
and choose the labeling so that
. With
, the mixed moments
are
|
(3)
|
where
is the rising factorial and
is a Lauricella
function of type
(Lakhani 2026). This labeling ensures
, within the absolute
convergence region of its power series. The moment of order zero is 1. For
the representation reduces to a Gauss hypergeometric
function, and for
to the Appell
hypergeometric function
.
Unlike the ordinary Dirichlet distribution,
unequal scales allow positive pairwise covariance
when
.