In the Gaussian polygenic models developed by Lande, the genetic contr
ibution to the variability of a phenotypic character is given by a sum
of allelic effects at several loci subject to mutations. The environm
ental contribution to that variability is the result of a random devia
tion, independent of the allelic effects, with a zero mean and a const
ant variance from generation to generation. The dynamics of the geneti
c variability is described by recurrence equations, from one generatio
n to the next, for the covariance matrix of the multinormal vector cor
responding to the allelic effects. These equations admit only one equi
librium and global convergence toward that equilibrium is checked by n
umerical iterations. In this article we present a Gaussian polygenic m
odel in which the selection forces (viability, preferential mating, fe
rtility) act generally on the vector of allelic effects, these being s
ubject to mutation, segregation and recombination. In order to study t
he convergence of variability, we establish a criterion for the conver
gence of the iterates of transformations on. semi-positive definite ma
trices tinder the condition that there exists a unique fixed point. Th
is criterion is essentially a concavity property which, combined with
a monotonicity property, has previously been wed by Karlin to demonstr
ate the convergence of variability in, Gaussian, phenotypic models wit
hout recombination or segregation. We show that the monotonicity prope
rty does not have to be assumed and we give a direct proof of the conv
ergence result under slightly weakened hypotheses. Finally, we show th
at this result applies to Gaussian polygenic models without difference
s between the sexes or without linkage between the loci. The robustnes
s of the results without these hypotheses and the rate of convergence
are studied by numerical iterations.