The use of Monte Carlo calculations in reactor criticality and shielding problems requires cross section data sets which are properties of the individual isotopes rather than group averaged sets. A major obstacle in containing such data entirely within a high speed computer memory has been the lack of a suitable method for producing such data sets in the unresolved resonance energy range. Up to now, two methods have been available:

  1. Generation of a point cross section data set based on a ladder of pseudoresolved resonances selected randomly from known average parameters and statistical laws.
  2. Generation of point cross sections during the Monte Carlo calculation, as needed, from stored average parameters.
The first method is hardly feasible in view of the enormous storage requirements while the second method would require excessive computation time in fast reactor calculations. A new method has been successfully applied to the analysis of fast critical assemblies in the VIM code. Cross section probability tables are appropriately distributed through the unresolved energy range of a given isotope. These tables consist of a probability distribution of cross sections to be used in an energy range surrounding the table energy. They are generated from point data sets obtained from ladders produced about a small energy range, sufficient to contain 50 to 100 resonances, insuring an adequate sampling of resonance interference and overlap effects while preventing significant variation in the energy dependent average parameters. The probability table method assumes that the resonance energies are sufficiently close that the neutron enters a resonance randomly, i.e. that the cross section seen by a neutron at one energy is in no way correlated with that at another energy. Cross sections are obtained rapidly from these tables during a Monte Carlo calculation by a random selection from the probability distribution described by the table assigned to the neutron energy, while storage requirements for a typical isotope are of the order of 1500 locations. The method has been thoroughly tested and appears to represent the unresolved region as well as the data permits while achieving computational efficiency in severely limited space.