function(k, n0, pstar, conf, rep)
{
	# last update 2/23/00
	# function to compute critical constants for comparisons with a standard
	# when mu is unknown and var-cov matrix satisfies  sphericity
	# k = number of systems, excluding baseline
	# n0 = initial sample size
	# pstar = single 1-alpha (PCS)
	# conf = confidence level on upper bounds for critical values
	# rep = number of reps to estimate critical values
	#
	# returns (g,h) with conf*100% upper confidence bounds
	#      (confidence on h is actually 1 - (1-conf)/2)
	#
	# first get h
	set.seed(13.)
	conf <- 1 - (1 - conf)/2
	Z <- matrix(rnorm(k * rep), ncol = k)
	Z0 <- matrix(rnorm(rep), ncol = 1.)
	Z0mat <- matrix(Z0, nrow = rep, ncol = k, byrow = F)
	C <- sqrt(matrix(rchisq(rep, (n0 - 1.) * k), ncol = 1.)/(k * (n0 - 1.))
		)
	Cmat <- matrix(C, nrow = rep, ncol = k, byrow = F)
	MVT <- (sqrt(1./2.) * Z - sqrt(1./2.) * Z0mat)/Cmat
	H <- sort(apply(MVT, 1., max))
	Hstar <- quantile(H, pstar)
	upper <- ceiling(pstar * rep + qnorm(conf) * sqrt(pstar * (1. - pstar) *
		rep) + 0.5)
	#
	#
	# then get g
	Hupper <- H[upper]
	Z <- matrix(rnorm(k * rep), ncol = k)
	Z0 <- matrix(rnorm(rep), ncol = 1.)
	Z0mat <- matrix(Z0, nrow = rep, ncol = k, byrow = F)
	C <- sqrt(matrix(rchisq(rep, (n0 - 1.) * k), ncol = 1.)/(k * (n0 - 1.))
		)
	Cmat <- matrix(C, nrow = rep, ncol = k, byrow = F)
	MVT <- (sqrt(1./2.) * Z - sqrt(1./2.) * Z0mat)/Cmat
	MVT[, 1.] <- MVT[, 1.] + Hupper
	G <- sort(apply(MVT, 1., max))
	Gstar <- quantile(G, pstar)
	upper <- ceiling(pstar * rep + qnorm(conf) * sqrt(pstar * (1. - pstar) *
		rep) + 0.5)
	Gupper <- G[upper]
	result <- cbind(c(Hstar, Hupper), c(Gstar, Gupper))
	result
}