minxent (version 0.01)

minxent.multiple: Minimum Cross Entropy Distribution under Multiple Constraints

Description

minxent.multiple estimates the MinxEnt distribution under given moment constraints.

Usage

"minxent"(q, G, eta, lambda)

Arguments

q

a priori distribution.

G

matrix of moment vector functions.

eta

vector of moment constraints.

lambda

initial points for langrangian multipliers.

Value

minxent.multiple returns an estimate of Lagrange multipliers and minimum cross entropy distribution under multiple constraints which is specified by user.

Warning

Since first Lagrange multiplies is a function of the others, there exists (m-1) constraints. (See. Kapur&Kesavan(1992) pp.44).

Details

MinxEnt distribution is obtained by Kullback's minimum cross entropy principles. This principle is introduced by Kullback (1957) which minimizes Kullback-Leibler divergence subject to given constraints. If a priori distribution is taken to be uniform distribution then minimizing Kullback-Leibler divergence is equivalent to maximizing Shannon's entropy subject to same given constraints. In the other words, in this special case MaxEnt distribution introduced by Jaynes (1957) is equivalent to MinxEnt distribution. For various application see Kapur&Kesavan(1992).

References

Jaynes, E. T. (1957). Information Theory and Statistical Mechanics. Physical Reviews, 106: 620-630. Kapur, J.N. and Kesavan, H.K.(1992), Entropy Optimization Principle with Applications, Academic Press. Kullback, S. (1959). Information Theory and Statistics. John Wiley, New York.

See Also

minxent.single

Examples

xi <- 1:6 
eta<-c(1,4,19)  #expected moment constraints 
q<-c(rep(1/6),6)  #a priori distribution 
G<-matrix(c(rep(1,6),xi,xi^2),byrow=TRUE,nrow=3) #matrix of moment vector function of observed data
minxent.multiple(q=q,G=G,eta=eta,c(0,0)) #estimates of lagrangian multipliers and MinxEnt distribution