A procedure is described that is used to restrain the iterate estimates of a dependent variable (such as the local neutron flux) when acceleration is applied. This scheme allows negative values but prevents extrapolation from positive values to a negative result.