newscaleunit_ function(S,c=log(10),k2=200,tol=1e-4,M=1e6,mu1pen=-0.5,mu2pen=0.5,sig2pen=0.75,pi1pen=0.5,pi2pen=0.5)
{

# Scale a data set so that the presumption of unit variance for
# the mixture components is tenable.  This can be done prior to
# invocation of  normal.S, so that the tests performed therein
# are meaningful. 
#
# More specifically, after centering the data, we apply the EM 
# algorithm (using the penalized maximum likelihood estimation
# framework of Chen, Chen, and Kalbfleisch) to the model 
# 		p_1 f(x | mu_1, sigma^2) + p_2 f(x | mu_2, sigma^2).  
# Then we scale the data using the resulting estimate of sigma^2.
#
# Required Input:
# 	S  is a vector of data
#
# Optional Inputs:
#     mu1pen, mu2pen  are multiplied by the sample standard
#        deviation to produce initial values.
#     sig2pen  is multiplied by the sample variance to produce 
#	   an initial value.
#     pi1pen, pi2pen  are initial values.
#     k2  is the maximum number of EM iterations allowed.
#     c, tol, M  are as described in  normal.S.
#
# Outputs:
# 	S  is a vector of scaled data


S_ S - mean(S)
n_ length(S)


mu1pen_ mu1pen*sqrt(var(S))
mu2pen_ mu2pen*sqrt(var(S))
sig2pen_ sig2pen*var(S)
nit2_ 0 

storage.mode(S)_"double"

storage.mode(mu1pen)_"double"
storage.mode(mu2pen)_"double"
storage.mode(sig2pen)_"double"
storage.mode(pi1pen)_"double"
storage.mode(pi2pen)_"double"

storage.mode(c)_"double"
storage.mode(M)_"double"
storage.mode(tol)_"double"
 


w_.Fortran("newscaleunit", 
		S=as.double(S),
	        n=as.integer(n), 	        
		  nit2=as.integer(nit2), 
		  c=as.double(c),
		  M=as.double(M),
		  k2=as.integer(k2),
		  tol=as.double(tol),
		pi1pen=as.double(pi1pen), 
		pi2pen=as.double(pi2pen), 
		mu1pen=as.double(mu1pen), 
		mu2pen=as.double(mu2pen),
		sig2pen=as.double(sig2pen))

S_ S/sqrt(w$sig2pen) 

return(S)
result
}