Professional Documents
Culture Documents
Gary M. Johnson
Joint Warfare Analysis Center
Dahlgren, Virginia
gjohnson@jwac.mil
November 7, 2007
Abstract
The type 1 and type 2 forms of the Dirichlet distribution have been discussed
for many years, as is evident in the work of Kotz, et al. [KBJ00]. Both types of
distributions have diverse areas of application, ranging from biopharmaceuticals,
genetics, forensic science, geology, pattern recognition, business, and economics.
The type 3 distribution is becoming an area of interest for research, but has not
received the same attention as the others. This report attempts to demonstrate the
existence of a general distribution from which types 1,2 and 3 originate and from
which a very broad class of Dirichlet distributions, as well as other well known
multivariate distributions, are found. This work is based on the work of McDonald
and Xu [MX95].
1
2
1 Introduction
The Dirichlet distribution is one of the most important multivariate distributions and
appears in many applications. Areas of application include order statistics; probabilis-
tic constrained programming models; project evaluation and review technique (PERT);
biopharmaceuticals; genetics and evolution; forensic science; geology and geochemistry;
pattern recognition; business and economics; political and social science; and artificial
intelligence and machine learning. With such a diverse array of areas that this distri-
bution is found, we wish to examine the common properties of the types of Dirichlet
distributions used, to discuss how they are related, to discover a broader class of distri-
butions within which such distributions exist, and to explore many of the fundamental
properties of this class of distributions.
Historically, the use of the expression “generalized Dirichlet distribution” has been used
extensively in describing a class of Dirichlet distributions that satisfy various correlational
properties among variables in the distribution (See, for example, the works of Connor
and Mosimann [CM69] and Wong [Won98].)
In a similar manner as McDonald and Xu, Section 5 lists several known multivariate
distributions that are defined using special parameter setting in GD. A taxonomy of
distributions is used to display the interrelationships of the distributions. A brief sum-
mary concludes this report.
The following special notation will be used throughout this presentation: We will use
the uppercase character to denote random vectors, such as X0 = (X1 , · · · , Xk ). This
will commonly be a vector of length k where k ≥ 1. The vector x, which will denote
an instance of the random vector X, will be defined as x0 = (x1 , · · · , xk ) where each
Pk Q
k
xi ≥ 0 and xi ≤ 1. For differentials of vector x, let dx = dxi . For a generic
i=1 i=1
vector parameter
³ ´ α where α0 = (α1 , · · · , αk ), let −α0 = (−α1 , · · · , −αk ), and α10 =
1 1 0
α1 , · · · , αk . For any vector w, let w = (w1 , · · · , wk ), unless otherwise noted. In
this presentation, we define a Dirichlet distribution to which we will assign a new type
3
designation; this will be the Dirichlet type c or generalized Dirichlet type c. The type
c refers to a vector and is emboldened since the distribution that it is assigned to is
defined distinctly by this vector. When the random vector X has the most general form
of Dirichlet distribution we will symbolize this as X ∼ GD(a, b, c, ν).
2 Definition
The generalized Dirichlet distribution is of the following form:
· ³ ´ai ¸νk+1 −1
Q
k Q
k P
k
Γ(ν+ ) |ai | yiai νi −1
1− (1 − ci ) ybii
i=1 i=1 i=1
GD(y | a, b, c, ν) = · ³ ´ai ¸ν+ (2.1)
Q
k+1 Q
k P
k
Γ(νi ) ai ν i
bi 1+ ci ybii
i=1 i=1 i=1
0
where a = (a1 , · · · , ak ) ∈ < , b = (b1 , · · · , bk ) ∈ <+k , c0 = (c1 , · · · , ck ) ∈ [0, 1]k ,
0 k
P
k+1
ν 0 = (ν1 , · · · , νk+1 ) ∈ <+k , with ν+ = νi . Throughout this presentation, we assume
i=1
that, for all i, νi > 0.
½ ³ ´ ai ¾
P
k
yi
The simplex S = yi | yi ≥ 0 for i = 1, · · · , k , (1 − ci ) bi ≤ 1 is the domain
i=1
of integration for GD. Throughout this presentation, since parameters can modify the
shape of the simplex S, the symbol S will be used without explicit reference to the vector
parameters a, b, or c.
Note 2.1. The generalized beta distribution, which is denoted as GB, is a special case
of the generalized Dirichlet distribution when k = 1. McDonald and Xu [MX95] provide
a complete examination of the properties of this distribution as well as its relationship
to numerous well-known continuous probability distributions.
3 Derivation of GD
We will show that GD is a probability distribution function by first deriving this dis-
tribution from the Dirichlet type 1 distribution (or equivalently from the Dirichlet type
2). This will form a distribution that will be called GD(a, b, c, ν) defined at a point
c = (c1 , · · · , ck ) where 0 ≤ ci ≤ 1 is a point in a unit hypercube. Using this distribution,
and assuming a = b = 1 for simplicity, we derive GD(1, 1, c(2) , ν) from GD(1, 1, c(1) , ν),
(1) (1) (2) (2)
defined at points c(1) = (c1 , · · · , ck ) and c(2) = (c1 , · · · , ck ), respectively. Lastly,
we derive this distribution from k + 1 independent gamma distributed random variables.
4
Assume that we have a Dirichlet type 1 distribution D1 (x | ν) or, equivalently, GD1 (x | 1, 1, ν).
Let X ∼ D1 (ν) with X0 = (X1 , · · · , Xk ). Define the spaces X and Y with X ∈ X and
Y ∈ Y, where Y is defined by the transformation T defined below. Our objective is
to determine the distribution of Y. Thus, suppose we apply the transformation T such
that
Xi
T : Yi = , for i = 1, · · · , k
P
k
1− ci Xi
i=1
Let y0 = (y1 , · · · , yk ) and Ik be the identity matrix of order k. The Jacobian is then
5
¯ ¯
¯ ∂(x1 , · · · , xk ) ¯
J = ¯¯ ¯
∂(y1 , · · · , yk ) ¯
à !−2k ¯ à ! ¯
k
X ¯ k
X ¯
¯ ¯
= 1+ ci yi ¯ 1+ ci yi Ik − c · y0 ¯
¯ ¯
i=1 i=1
à k
!−2k à k
!k−1
X X
= 1+ ci yi 1+ ci yi
i=1 i=1
à k
!−(k+1)
X
= 1+ ci yi .
i=1
µ ¶
P
k à !−1 νi −1
Γ νi k
Y k
X
i=1 yi 1 +
GD(y | 1, 1, c, ν) = · ci yi
Q
k
Γ(νi ) i=1 i=1
i=1
à !−1 νk+1 −1 à !−(k+1)
k
X k
X
× 1 − yi 1+ ci yi 1+ ci yi
i=1 i=1
µ ¶
P
k
à !−ν+ " #νk+1 −1
Γ νi k
Y k
X k
X
i=1
= · yiνi −1 1+ ci yi 1− (1 − ci )yi
Q
k
Γ(νi ) i=1 i=1 i=1
i=1
(3.1)
½ ¾
P
k
where S = yi | yi ≥ 0 for i = 1, · · · , k , (1 − ci )yi ≤ 1 .
i=1
³ ∗ ´ ai
y
Using the variable transformation yi = bii for i = 1, · · · , k, the probability distribu-
tion Equation 3.1 is written in the general Dirichlet distribution form GD(y∗ | a, b, c, ν)
shown in Equation 2.1.
6
Let Y ∼ GD(1, 1, c(1) , ν) with Y0 = (Y1 , · · · , Yk ). Assume the general Dirichlet dis-
(2) (1)
tribution as defined in Equation 3.1, with c(1) = c. Let c∗i = ci − ci , where
(1) (2)
ci ∈ [0, 1] and ci ∈ [0, 1], for i = 1, · · · ¯, k. It is required
¯ that −1 ≤ c∗i ≤ 1, giv-
(2) (1) ¯ (2) (1) ¯
ing us −1 ≤ ci − ci ≤ 1 or, equivalently, ¯ci − ci ¯ ≤ 1.
µ ¶−(k+1)
P
k
The Jacobian of this transformation is J = 1+ c∗i wi . Using this, the general
i=1
Dirichlet distribution is
7
GD(w | 1, 1,c(2) , ν)
µ k ¶ νi −1 −ν+
P
Γ νi k
Y Xk
i=1 wi (1) wi
= k · 1 + ci
Q P
k P
k
Γ(νi ) i=1 1 + c∗i wi i=1 1+ c∗i wi
i=1 i=1 i=1
νk+1 −1
à !−(k+1)
Xk ³ ´ wi k
X
(1)
× 1 − 1 − ci 1+ c∗i wi
P
k
i=1 1+ c∗i wi i=1
i=1
µ ¶
P
k
" #−ν+ ( )νk+1 −1
Γ νi k
Y k ³
X ´ k h
X ³ ´i
i=1 (1) (1)
= · wiνi −1 1+ ci + c∗i wi 1+ 1 − ci + c∗i wi
Q
k
Γ(νi ) i=1 i=1 i=1
i=1
µ ¶
P
k
à !−ν+ " #νk+1 −1
Γ νi k
Y k
X k ³
X ´
i=1 (2) (2)
= · wiνi −1 1+ ci wi 1− 1− ci wi
Q
k
Γ(νi ) i=1 i=1 i=1
i=1
(3.2)
½ k ³ ´ ¾
P (2)
defined on the simplex S = wi | wi ≥ 0 for i = 1, · · · , k , 1− ci wi ≤ 1 .
i=1
Consequently, we see that the general Dirichlet type c(2) distribution GD(·|1, 1, c(2) , ν)
is derivable at all points c(2) ∈ [0, 1]k from any Dirichlet type c(1) distribution.
Example 3.1. Let c(1) = 0 and c(2) = 1 so that c∗ = 1. Then our transformation is
from the Dirichlet type 1 to Dirichlet type 2. Similarly, if we let c(1) = 1 and c(2) = 0
so that c∗ = −1, then our transformation is from the Dirichlet type 2 to Dirichlet type
1.
Example 3.2. If we let c(1) = 0 and c(2) = c so that c∗ = c, then the transformation
Tc allows a derivation of the distribution GD(·|1, 1, c, ν) from the distribution D1 , as
was demonstrated in Section 3.1. A similar result follows when we let c(1) = 1 and
c(2) = c so that c∗ = c − 1. In this case, the transformation allows a derivation of the
distribution GD(·|1, 1, c, ν) from the inverse Dirichlet distribution D2 . The distribution
D2 is discussed further in Section 5.6.
8
Definition 3.1. To distinguish GD(· | a, b, c, ν) from GD(· | 1, 1, c, ν), we will call the
latter form Dirichlet type c.
Note that when Dirichlet type 1 is renamed Dirichlet type 0 where 00 = (0, · · · , 0) and
Dirichlet type 2 is renamed Dirichlet type 1 where 10 = (1, · · · , 1), then this new vector-
designator is more descriptive in notation than the currently assigned notations of types
1 and 2. Likewise, the Dirichlet type 3 notation can now be renamed Dirichlet type 12
0 ¡ ¢
where 12 = 21 · · · 12 . In general, the vector-designator for the general Dirichlet type c
where c0 = (c1 , · · · , ck ) is used to define the general Dirichlet distribution.
Suppose that the random variable Zi has a gamma distribution with parameter νi , or
Zi ∼ Γ(νi ), for i = 1, · · · , k + 1 and that the transformation S is defined as
Xi = k+1
P
Zi
for i = 1, · · · , k
Zi
i=1
S:
P
k+1
Xk+1 = Zi
i=1
Define the set Z with random variables Zi ∈ Z. Note that X0 = (X1 , · · · , Xk ) has the
Dirichlet type 1 distribution, or X ∼ D1 (ν). ( Kotz, et al [KBJ00] Chapter 40, Section
9
The transformation T ◦ S is defined using the transformations T from Section 3.1 and S:
Zi
P
k+1
Zj
Zi
Y =
j=1
= for i = 1, · · · , k
i
P
k
P Zi
k+1 Zk+1 + (1 − cj )Zj
1− ci k+1 j=1
P
T ◦S : i=1 Zj
j=1
k+1
X
Y = Zi
k+1
i=1
The transformations S, T and the composite transformation T ◦S are shown in the figure
below:
X
~? @@@
S ~~~ @@T
~~~ @@
~ Â
Z /Y
T ◦S
Zi
Example 3.3. When c0 = (0, · · · , 0), Yi = P
k+1
for i = 1, · · · , k, which corresponds
Zj
j=1
with the Dirichlet type 1 random variable (See Kotz, et al [KBJ00], Chapter 49, Section
1).
à k
!k−1
k
yk+1 X
=µ ¶2k 1+ ci yi
P
k
i=1
1+ ci yi
i=1
k
yk+1
=µ ¶k+1 .
P
k
1+ ci yi
i=1
−ν+ νk+1 −1
k
Y k
X k
X
Γ (ν+ ) ν −1
= k+1 yj j 1 + cj yj 1 − (1 − cj ) yj .
Q
Γ (νj ) j=1 j=1 j=1
j=1
(3.6)
4 Properties of GD
In order to characterize GD, we determine its moment generating function E (Y1r1 · · · Ykrk ).
From this, we then demonstrate several well known special cases of the moment gener-
ating function. Following this, we derive the marginal distribution for GD along with
several cases of special interest.
The moment generating function for GD is developed by the use of the Lauricella hyper-
geometric function type D and the Gauss hypergeometric function. See the monograph
Exton [Ext76] for a thorough examination of hypergeometric functions used in this sec-
tion.
Using multivariate expected value operations with GD, we get the following:
Z Z Y
k
E (Y1r1 · · · Ykrk ) = ··· yiri GD(y | a, b, c, ν)dy
S i=1
· ³ ´ai ¸νk+1 −1
Q
k Q
k Pk
Z Z Γ(ν+ ) |ai | yiai νi +ri −1 1 − (1 − ci ) ybii
i=1 i=1 i=1
= ··· · ³ ´ai ¸ν+ dy
Q
k+1 Q k
ai νi Pk
yi
S Γ(νi ) bi 1+ ci bi
i=1 i=1 i=1
(4.1)
k k Z Z Yk
à k
!νk+1 −1
Γ(ν+ ) Y ri Y r
− νi + ai
r
νi + ai −1 X
= k+1 bi (1 − ci ) i ··· wi i
1− wi
Q
Γ(νi ) i=1 i=1 S i=1 i=1
i=1
" k µ ¶ #−ν+
X −ci
× 1− wi dw
i=1
1 − ci
(4.3)
½ ¾
P
k
where S = wi | wi ≥ 0, i = 1, · · · , k, wi ≤ 1 .
i=1
(k)
Using Lauricella function FD , then Equation 4.3 becomes
Q
k³ ´
k
Y k
Y Γ(νk+1 ) Γ νi + arii
Γ(ν+ ) r
− νi + ai i=1
bri i (1 − ci ) i µ ¶
Q
k+1 P
k
ri
Γ(νi ) i=1 i=1 Γ ν+ + ai
i=1 i=1
à k
!
(k) r1 rk X ri −c1 −ck
× FD ν+ , ν1 + , · · · , νk + ; ν+ + ; ,··· ,
a1 ak a
i=1 i
1 − c1 1 − ck
(4.4)
Q
k ³ ´
k
Y Γ(ν+ ) Γ νi + arii
i=1
= bri i k µ ¶
Q P
k
ri
i=1 Γ (νi ) Γ ν+ + ai
i=1 i=1
à k k
!
(k)
X ri r1 rk X ri
× FD , ν1 + , · · · , νk + ; ν+ + ; c1 , · · · , ck .
a
i=1 i
a1 ak a
i=1 i
14
Since
à k k
!
(k)
X ri r1 rk X ri
FD , ν1 + , · · · , νk + ; ν+ + ; c1 , · · · , ck
a
i=1 i
a1 ak a
i=1 i
µ ¶
Pk
ri
Γ ν+ + ai Z1 P ri Pk
r
k ν+ − ci νi + ai −1
i=1 −1
= µ k ¶ u i=1 ai
(1 − u) i=1 i du
P ri (4.5)
Γ ai Γ (ν+ ) 0
i=1
µ ¶ · ³ ´¸
P
k
ri P
k
ri
Γ ν+ + ai Γ ν+ − ci νi + ai
i=1 i=1
= · ³ ´ ¸.
Γ (ν+ ) P
k
ri
Γ (1 − ci ) νi + ai + νk+1
i=1
we have
Q
k ³ ´ · P
k ³ ´¸
ri ri
k
Y Γ νi + ai Γ ν+ − ci νi + ai
i=1 i=1
E (Y1r1 , · · · , Ykrk ) = bri i · k ³ ´ ¸. (4.6)
Q
k P ri
i=1 Γ (νi ) Γ (1 − ci ) νi + ai + νk+1
i=1 i=1
Example 4.1. If we let ri = 0 for all i = 1, · · · , k, then from Exton ([Ext76], Equations
2.3.5 and 2.3.6), using Equation 4.4, we get
Z Z
(k)
··· GD(y | a, b, c, ν)dy = FD (0, ν1 , · · · , νk ; ν+ ; c1 , · · · , ck )
S
Z Z Y
k
à k
!νk+1 −1
Γ(ν+ ) X
= k+1 ··· ziνi −1 1− zi dz
Q
Γ(νi ) S i=1 i=1
i=1
Q
k+1
Γ(νi )
Γ(ν+ ) i=1
= k+1
Q Γ(ν+ )
Γ(νi )
i=1
=1
(4.7)
Q
k+1
Γ(νi )
i=1
where the right-hand multiple integral has measure , being a Dirichlet type 1
Γ(ν+ )
probability distribution function, defined in Section 5.1.
15
Example 4.2. If we set ci = 0 for all i = 1, · · · , k, then Equation 4.4 can be written as
Q
k ³ ´
ri
k
Y Γ(ν+ ) Γ νi + ai
i=1
E (Y1r1 · · · Ykrk ) = bri i ·k+1 ³ ´¸
Q
k P ri
i=1 Γ(νi )Γ νi + ai
i=1 i=1
³ ´
Q
k Γ νi + arii (4.8)
Yk i=1 Γ(νi )
= bri i ·k+1 ³ ´¸
P ri
i=1 Γ νi + ai
i=1
Γ(ν+ )
where rk+1 = 0. This is the moment generating function for the generalized Dirichlet
type 1 probability distribution function, defined in Section 5.1. When ai = bi = 1
for all i = 1, · · · , k, we have the moment generating function for the Dirichlet Type 1
probability distribution function (See Kotz, et al. [KBJ00], p. 488).
Example 4.3. If we set ci = 1 for all i = 1, · · · , k, then Equation 4.4 can be written as
Q
k ³ ´
ri
k
Y Γ(ν+ ) Γ νi + ai
i=1
E (Y1r1 · · · Ykrk ) = bri i · ´¸
Q
k P³
k+1
ri
i=1 Γ(νi )Γ ν+ + ai
i=1 i=1
à k k
!
(k)
X ri r1 rk X ri
× FD , ν1 + , · · · , νk + ; ν+ + ; 1, · · · , 1
a
i=1 i
a1 ak a
i=1 i
Q
k ³ ´ µ P
k
¶
ri ri
k
Y Γ νi + ai Γ νk+1 − ai
= bri i i=1 i=1
Q
k+1
i=1 Γ(νi )Γ(νk+1 )
i=1
³ ´
Q
k Γ νi + arii
k
Y i=1 Γ(νi )
= bri i
Γ(νk+1 )
i=1 µ ¶
Pk
ri
Γ νk+1 − ai
i=1
16
P
k
ri
where νk+1 − ai > 0. This is the moment generating function for the generalized
i=1
Dirichlet type 2 probability distribution function, defined in Section 5.2. In particular, if
we set ai = bi = 1 for all i = 1, · · · , k, then we have the moment generating function for
the Dirichlet type 2 probability distribution function. (See Kotz, et al. [KBJ00], p. 492
for more details).
1
Example 4.4. If we set ci = 2 for all i = 1, · · · , k, then Equation 4.4 can be written as
Q
k
k
Y Γ(ν+ ) Γ (νi + ri ) Γ(νk+1 )
i=1
E (Y1r1 · · · Ykrk ) = bri i k+1 · k ¸
Q P
i=1 Γ(νi )Γ (νi + ri ) + νk+1
i=1 i=1
(4.9)
à k k
!
(k)
X ri r1 rk X ri 1 1
× FD , ν1 + , · · · , νk + ; ν+ + ; ,··· ,
a
i=1 i
a1 ak a 2
i=1 i
2
This is defined as the moment generating function for the generalized Dirichlet type 3
distribution, defined in Section 5.3. Using Equation 4.9, if we set ai = 1 and bi = 12 , for
i = 1, · · · , k, then
Q
k
k µ ¶ri
Y Γ(ν+ ) Γ (νi + ri ) Γ(νk+1 )
1 i=1
E (Y1r1 · · · Ykrk ) = · k ¸
2 Q
k+1 P
i=1 Γ(νi )Γ (νi + ri ) + νk+1
i=1 i=1
à k k
!
(k)
X X 1 1
× FD ri , ν1 + r1 , · · · , νk + rk ; ν+ + ri ; , · · · ,
i=1 i=1
2 2
P
k (4.10)
− ri Q
k
2 Γ(ν+ )
i=1 Γ(νi + ri )
i=1
= k · k ¸
Q P
Γ(νi )Γ (νi + ri ) + νk+1
i=1 i=1
à k k k
!
X X X 1
× 2 F1 ri , (νi + ri ); ν+ + ri ;
i=1 i=1 i=1
2
using results from Exton ([Ext76], p. 288, Equation A.2.10) and where 2 F1 is the Gauss
hypergeometric function.
17
Q
k
à !
2−νk+1 Γ(ν+ ) Γ(νi + ri ) Xk
r1 rk i=1 1
E (Y1 · · · Yk ) = k · k ¸ 2 F1 νk+1 , ν+ ; ν+ + ri ;
Q P 2
Γ(νi )Γ (νi + ri ) + νk+1 i=1
i=1 i=1
k k
à k
!νk+1 −1 " k µ ¶ #−ν+
Γ(ν+ ) Y −ν
Y X X −ci
f (w, c, ν) = k+1 (1 − ci ) i wiνi −1 1− wi 1− wi .
Q 1 − ci
Γ(νi ) i=1 i=1 i=1 i=1
i=1
(4.11)
ci (1) c0i
Let c0i = − and ci = − P
m where 1 ≤ m < k and ci < 1 for i = 1, · · · , m.
1 − ci
1− c0i wi
i=1
µ ¶− ν+
P
k
From Equation 4.11, the expression 1− c0i wi can be written as
i=1
18
− ν +
à !− ν+ P
k
m c0i wi
X i=m+1
1− c0i wi 1 − P
m
i=1 1− c0i wi
i=1
à m
!− ν+ Ã k
!− ν+
X X (1)
= 1− c0i wi 1− ci wi
i=1 i=m+1
à !− ν+
m
X X X
= 1− c0i wi (ν+ , lm+1 )(ν+ + lm+1 , lm+2 ) · · · ν+ + lj , lk
i=1 lm+1 , ··· , lk m+1 ≤ j < k
·³ ´lm+1 ³ ´l ¸ lm+1
(1) (1) k wm+1 wlk
× −cm+1 · · · −ck ··· k
lm+1 ! lk !
à !− ν+ à !·
m
X X k
X ³ ´lm+1 ³ ´lk ¸ wlm+1 wklk
(1) (1) m+1
= 1− c0i wi ν+ , li −cm+1 ··· −ck ···
i=1 i=m+1
lm+1 ! lk !
lm+1 , ··· , lk
Γ(a + n) P
where (a, n) = , with a > 0 and n ∈ Z and where is the multiple
Γ(a) lm+1 , ··· , lk
sum over lm+1 , · · · , lk , where 0 ≤ lj < ∞ for j = m + 1, · · · , k. Then the marginal
distribution of GD, denoted GD(m) , for variables (w1 , · · · , wm ), is
GD(m) (w | 1, 1, c, ν)
k Z Z Yk
à k
!νk+1 −1
Γ(ν+ ) Y X
= k+1 (1 − ci )−νi
··· wiνi −1 1− wi
Q
Γ(νi ) i=1 S i=1 i=1
i=1
à !− ν+ à !·
m
X X k
X ³ ´lm+1 ³ ´lk ¸ wlm+1 wklk
(1) (1) m+1
× 1− c0i wi ν+ , li −cm+1 ··· −ck ··· dw
i=1 i=m+1
lm+1 ! lk !
lm+1 , ··· , lk
Since
Z Z Y
m m
à k
!νk+1 −1
Γ(ν+ ) Y X
··· wiνi −1 wiνi +li −1 1− wi dw
Q
k+1
Γ(νi ) S i=1 i=1 i=1
i=1
Q
k
à !Pk+1 Pk
Γ(ν+ ) Γ(νi + li ) m m i=m+1 νi + i=m+1 li −1
i=m+1
Y X
νi −1
= k+1 µ k+1 ¶ wi 1 − wi
Q P Pk
Γ(νi )Γ νi + li i=1 i=1
i=1 i=m+1 i=m+1
Q
k
à ! Pk+1 Pk
Γ(ν+ )
(νi , li ) m m νi + i=m+1 li −1
i=m+1
Y X i=m+1
νi −1
= m µ k+1 ¶ µ k+1 ¶ wi 1− wi
Q P P P
k
Γ(νi )Γ νi Γ νi , li i=1 i=1
i=1 i=m+1 i=m+1 i=m+1
we get
GD(m) (w | 1, 1, c, ν)
k
à m
!− ν+ m
à m
!Pk+1
j=m+1 νj −1
Γ(ν+ ) Y X Y X
= m µ k+1 ¶ (1 − ci )−νi 1+ c0i wi wiνi −1 1− wi
Q P
Γ(νi )Γ νi i=1 i=1 i=1 i=1
i=1 i=m+1
P
m
li
µ ¶ 1− wi
P
k Q
k − c0i i=1
ν+ , li (νi , li ) P
m
X k
Y 1+ c0i wi
i=m+1 i=m+1 i=1
× µ k+1 ¶
P Pk li !
lm+1 ,··· ,lk νi , li i=m+1
i=m+1 i=m+1
(4.12)
Applying the transformation T from Equation 4.2 we can write this expression as
20
GD(m) (y | a, b, c, ν)
Q
m
Γ(ν+ ) |ai |
i=1
= m µ k+1 ¶ m
Q P Q ai νi
Γ(νi )Γ νi bi
i=1 i=m+1 i=1
" P
k
Y m
X µ ¶ai #− ν+ Y
m Xm
" µ ¶ai # k+1 i=m+1 νi −1
−νi yi νi −1 y i
× (1 − ci ) 1+ ci yi 1− (1 − ci )
i=m+1 i=1
bi i=1 i=1
bi
P
m ³ ´ ai Pm ³ ´ ai
yi yi
k+1
X 1 − (1 − ci ) 1 − (1 − ci )
(k−m)
bi bi
× FD ν+ , νm+1 , · · · , νk ; νi ; − ck
0 i=1
³ ´ , · · · , − cm+1
0 i=1
³ ´
P
m a i Pm a i
i=m+1 1+ ci ybii 1+ ci ybii
i=1 i=1
(4.13)
½ ³ ´ ai ¾
P
m
yi
Equation 4.13 is defined on the simplex S = yi | yi ≥ 0 for i = 1, · · · , m , (1 − ci ) bi ≤1 ,
i=1
with c0 = (c1 , · · · , cm ) ∈ [0, 1]m and ci < 1, for m + 1 ≤ i ≤ k.
(m)
≡ GD1 (y | a, b, 0, ν)
( m µ ¶a )
X yi i
S= yi | yi ≥ 0 for i = 1, · · · , m , ≤1 (4.15)
i=1
bi
where ba1 1 > y1a1 ≥ 0. The function GB1 is the generalized beta type 1 distribution
function, defined by McDonald and Xu [MX95], Equation 2.1.
1
Example 4.7. If we let bi = ci = 2 and ai = 1 for i = 1, · · · , m, then
µ ¶
1 1
GD(m) y | 1, , , ν
2 2
Pk
µ ¶Pk+1
j=m+1 νj −1
Q
m Pm
Γ(ν+ ) 2 i=m+1 νi
1−yiνi −1
yi
i=1 i=1
= µ k+1 ¶ m µ ¶ν+
Q
m P Q Pm
Γ(νi )Γ νi 1+ yi
i=1 i=m+1 i=1 i=1
P
m P
m
k+1
X 1− yi 1− yi
(k−m)
× FD ν+ , νm+1 , · · · , νk ; νi ; − i=1 ,··· ,− i=1
Pm Pm
i=m+1 1+ yi 1+ yi
i=1 i=1
Pk
µ ¶Pk+1
j=m+1 νj −1
Q
m Pm
Γ(ν+ ) 2 i=m+1 νi yiνi −1 1 − yi
i=1 i=1
= µ k+1 ¶ m µ ¶ν+
Q
m P Q Pm
Γ(νi )Γ νi 1+ yi
i=1 i=m+1 i=1 i=1
P
m
k
X k+1
X 1− yi
(k−m)
× 2 F1 ν+ , νi ; νi ; − i=1 .
Pm
i=m+1 i=m+1 1+ yi
i=1
(4.16)
This corresponds to the result of Cardenõ, et al., [CNS05] in which the marginal distri-
bution GD(m) is shown to not be in Dirichlet type 3 family of distributions.
22
µ ¶νk+1 −1
Q P
Γ (ν+ ) yiνi −1 1 − yi
i∈T ∪T 0 i∈T 0
GD(y | 1, 1, c, ν) = µ ¶ν+
Q P
Γ (νi ) Γ (νk+1 ) 1 + yi
i∈T ∪T 0 i∈T
½ ¾
P
defined over the simplex S = yi | yi ≥ 0 for i ∈ T ∪ T 0 and yi ≤ 1 . This proba-
i∈T 0
bility distribution function will be called the mixed type 1 - type 2 Dirichlet distribution
function since it combines properties of both Dirichlet type 1 and Dirichlet type 2. This
is easily determined to be true through the following two examples:
Z∞ Z∞ Y
f (yT 0 ) = ··· GD(y | 1, 1, c, ν) dyi
0 0 i∈T
µ ¶
P Ã !νk+1 −1
Γ ν+ − νi Y X
i∈T νi −1
= Q yi 1− yi .
Γ (νi ) Γ (νk+1 ) 0 0i∈T i∈T
i∈T 0
½ ¾
0
P
This is the Dirichlet type 1 defined on S1 = yi | yi ≥ 0 for i ∈ T and yi ≤ 1 .
i∈T 0
In a similar manner as just shown, we can find the distribution for yT = {yi | i ∈ T }:
Z Z Y
g(yT ) = ··· GD(y | 1, 1, c, ν) dyi
S1 i∈T 0
Q
yiνi −1
Γ (ν+ )
= µ ¶ µ i∈T ¶ν+ .
Q P P
Γ (νi ) Γ ν+ − νi 1+ yi
i∈T i∈T i∈T
From the prior two examples we observe that g(yT ) · f (yT 0 ) = GD(y|1, 1, c, ν), where c
is suitably chosen from {0, 1}k .
For more information on many of the distributions listed, see Kotz, et al. [KBJ00].
GD1 (y | a, b, ν) = GD(y | a, b, 0, ν)
· ¸
k ³ ´ai νk+1 −1
Q
k Q
k P yi
Γ(ν+ ) |ai | yiai νi −1 1− bi (5.1)
i=1 i=1 i=1
=
Q
k+1 Q
k
Γ(νi ) bai i νi
i=1 i=1
k ³ ´ai
P yi
where yi > 0 for i = 1, · · · , k, and bi ≤ 1.
i=1
24
1. Dirichlet type 1
Using the generalized Dirichlet type 1, Equation 5.1, when ai = bi = 1 for i =
1, · · · , k, then the Dirichlet type 1 distribution (denoted D1 ) is written
D1 (y | ν) = GD1 (y | 1, 1, ν)
µ ¶νk+1 −1
Q
k P
k
Γ(ν+ ) yiνi −1 1− yi (5.2)
i=1 i=1
=
Q
k+1
Γ(νi )
i=1
P
k
where yi ≥ 0 for i = 1, · · · , k, and yi ≤ 1.
i=1
P
k
where yi ≥ 0 for i = 1, · · · , k, and yi−1 ≤ 1.
i=1
βi for i = 1, · · · , k so that b∗ =
a
Let β = (β1 , · · · , βk ). Substituting bi = νk+1
i
µ 1 1
¶
a1 ak
νk+1 β1 , · · · , νk+1 βk in Equation 5.1, we get
f (y | a, β, ν) = GD(y | a, b∗ , 0, ν)
· k ³ ´¸νk+1 −1
Q
k Q
k P yi i
a
Γ(ν+ ) |ai | yiai νi −1 1− a
νk+1 βi i (5.4)
i=1 i=1 i=1
=
Q
k+1 Q
k
Γ(νi ) νi
νk+1 βi ai νi
i=1 i=1
k ³
P a ´
yi i
where yi ≥ 0 for i = 1, · · · , k, and a
νk+1 βi i
≤ 1.
i=1
Then the independent generalized Gamma is given by
a i
y
Yk ai νi −1 − βii
|a |
i iy e
IGG(y | a, ν, β) = lim f (y | a, β, ν) = ai νi
νk+1 →∞
i=1
Γ(ν i )βi
where yi ≥ 0 .
25
σ12 0
..
where 0 < yi < ∞ for i = 1, · · · , k and where Σ = . . Note that
0 σk2
this is defined only for positive variables.
GD2 (y | a, b, ν) = GD(y | a, b, 1, ν)
Q
k Q
k
Γ(ν+ ) |ai | yiai νi −1
i=1 i=1 (5.6)
= · ¸
Q
k+1 Q
k Pk ³ ´ai ν+
yi
Γ(νi ) bai i νi 1 + bi
i=1 i=1 i=1
1. Dirichlet Type 2
When ai = bi = 1 for i = 1, · · · , k, then the Dirichlet type 2, more commonly
referred to as the inverse Dirichlet distribution (denoted D2 ) is defined as
D2 (y | ν) = GD2 (y | 1, 1, ν)
Q
k
Γ(ν+ ) yiνi −1
i=1 (5.7)
= µ ¶ν+
Q
k+1 Q
k P
k
Γ(νi ) 1+ yi
i=1 i=1 i=1
Γ(νi ) Γ 2
i=1 i=1
Γ( k+1
2 )
M C(y) = k+1 £ ¤ k+1 (5.9)
π 2 1 + ( y21 )2 + · · · + ( y2k )2 2
µ ¶
P
k Q
k Q
k
(5.10)
Γ `i + a θi`i yi`i −1
i=1 i=1 i=1
= .
¶ P `i +a
k
µ
Q
k P
k i=1
Γ(a) Γ(`i ) 1 + θi y i
i=1 i=1
M L(y | a, θ) = GM L(y | a, θ, 1)
Q
k
Γ(k + a) θi
i=1 (5.11)
= µ ¶k+a
P
k
Γ(a) 1 + θi y i
i=1
where 0 < yi < ∞ . Nayak [Nay87] studies the multivariate Lomax distribution
with its generalization and demonstrates its relationship to multivariate f, multi-
variate Pareto Type 2, and multivariate Burr.
4. Multivariate f
Using the generalized multivariate Lomax distribution, if we let θi = a`ii for all
i = 1, · · · , k, then the multivariate f distribution (denoted M F ) is defined as
M F (y | `, a, a) = GM L(y | a, θ, `)
µ ¶ k ³ ´ k
P
k Q `i `i Q `i −1
Γ `i + a ai yi (5.12)
i=1 i=1 i=1
=
¸ P `i +a
k
· k ³ ´
Q
k P `i i=1
Γ(a) Γ(`i ) 1 + ai yi
i=1 i=1
where 0 < yi < ∞. For further information on the relationship between multivari-
ate Lomax and multivariate f, see Nayak [Nay87], p.176.
5. Multivariate Log-Logistic
Using the multivariate Lomax distribution Equation 5.11, when we set a = 1, then
the multivariate log-logistic distribution (denote MLL) is defined as
µ k ¶
Q
θi k!
i=1
M LL(y | θ) = µ ¶k+1 (5.13)
Pk
1+ θi y i
i=1
Q
k
Γ(k + a) ci yi ci −1
i=1
M B(y | a, c, d) = ci a+k
· ³ ´ 1 ¸ ci
Q
k
1c i P
k
yi
Γ(a) di 1 + ³ ´ 1
i=1 i=1 c
1 i
di (5.14)
Q
k
a(a + 1) · · · (a + k − 1) di ci yici −1
i=1
= · ¸a+k
P
k
1+ di yici
i=1
where 0 < yi < ∞ . The distribution M P2 is derivable from the inverted Dirichlet
distribution Equation 5.7 when we set νi = 1 for i = 1, · · · , k and νk+1 = a, as
well as from the multivariate Lomax distribution Equation 5.11 when θi = 1 for
i = 1, · · · , k. For further information on the multivariate Pareto distribution see
Mardia [Mar62].
P
k ³ ´ ai
1 yi
where yi ≥ 0 for i = 1, · · · , k, and 2 bi ≤ 1.
i=1
1. Dirichlet Type 3
In particular, if we set ai = 1 and bi = 12 , then the Dirichlet Type 3 distribution
(denoted D3 ) is defined as
P
k
where 0 < yi , νi > 0 for all i, yi < 1 and k ≥ 1. This distribution is the
i=1
multivariate generalization of the beta type 3 distribution, denoted B3 (See Car-
denõ, et al. [CNS05]). In particular, when k = 1, D3 (y | ν) = B3 (y | ν1 , ν2 ). For
additional information on B3 and D3 , see Cardenõ, et al., [CNS05].
P
k
1
where 0 < yi < ∞ for i = 1, · · · , k and yi ≤ 1.
i=1
30
6 PDF Taxonomy
A taxonomy is provided to organize the classes of multivariate distributions discussed
in Section 5 and illustrate the relationships between the three commonly used Dirich-
let distributions and other common multivariate distributions. The general Dirichlet
distribution (GD) is defined with the largest number of parameters, 4k + 1, and is de-
picted at the center. Distributions that are one step from GD have 3k + 1 parameters.
Distributions that are two or more steps from GD have less than 3k + 1 parameters.
GD( 2.1)
ggg
ggggggg
g
gggg
gggg
gs ggg ²
GD1 ( 5.1) GD2 ( 5.6) WW
qq OOO WWWW
qq OOO WWWWW
q OOO WWWWW
qqq OOO WWWWW
xqqq ² ² ' WW+
ID1 ( 5.3) D1 ( 5.2) GM L( 5.10) D2 ( 5.7) GM C( 5.8)
nnnn
nn
nnn
² wnnn ² ²
IG( 5.4) M F ( 5.12) M L( 5.11) M C( 5.9)
n OOO
nnnnn OOO
OOO
nnnn OOO
² wnn ² ' ²
IN ( 5.5) M LL( 5.13) M B( 5.14) M P2 ( 5.15)
Although numerous distributions have been identified and located in this taxonomy, it
is uncertain that it is complete.
7 Conclusion
By demonstrating that the Dirichlet distribution encompasses a much broader class
of distributions than has been shown to date allows us an opportunity to extend our
knowledge of this distribution. From what has been provided in this report, we have
31
seen that the generalized Dirichlet distribution GD includes a wide class of well-known
distributions as special cases. We have also found that it is consistent with the other
Dirichlet distributions when selected parameters are used.
The down side of this work is that marginal distributions are found to not be in the
same class of distributions as the original distribution. This may limit the possibility
of achieving useful results in such areas as conditional distributions in the general case.
This subject needs to be further investigated before making any further claims.
The GD distribution has been shown to be derivable from gamma and beta distribu-
tions; a strong possibility exists for this distribution to be extended in several areas of
investigation, including:
2. Based on the multitude of applications that have appeared in such areas as business,
economics, social science, biological science, and others, it is essential that this new
distribution be demonstrated in similar applications.
3. Developing further results that rely on the use of beta functions. This includes
extending the results to concepts of neutrality with applications to the generalized
Dirichlet distribution defined by Connor and Mosimann.
4. Since this work introduces us to a new type of Dirichlet distribution, we see the
possibility for extending this work to include, at the minimum the following items:
References
[Ait58] A. C. Aitken. Determinants and Matrices. Oliver and Boyd, 1958.
[MX95] James B. McDonald and Yexiao J. Xu. A generalization of the beta distribution
with applications. Journal of Econometrics, 66:133–152, 1995.
[Tak65] K. Takahasi. Note on the multivariate Burr’s distribution. The Annals of the
Institute of Statistical Mathematics, 17:257–260, 1965.
[Won98] T. T. Wong. Generalized Dirichlet distributions in Bayesian analysis. Applied
Mathematics and Computation, 97:165–181, 1998.