VIM uses standard Monte Carlo methods for particle tracking with several optional variance-reduction techniques. These include splitting/Russian roulette, non-terminating absorption with nonanalog weight cutoff energy. The keff is determined by the optimum linear combinations of two of the three eigenvalue estimates - analog, collision, and track length. Resonance and smooth cross sections are specified pointwise with linear - linear interpolation, frequently with many thousands of energy points. Unresolved resonances are described by the probability table method, which allows the statistical nature of the evaluated resonance cross sections to be incorporated naturally into the representation of self-shielding effects. Neutron interactions are elastic, inelastic and thermal scattering, (n,2n), fission, and capture, which includes (n,gamma), (n,p), (n,alpha), etc. Photon interaction data for pair production, coherent and incoherent scattering, and photoelectric events are taken from MCPLIB. Trajectories and scattering are continuous in direction, and anisotropic elastic and discrete level inelastic neutron scattering are described with probability tables derived from evaluated nuclear data. VIM has an automatic restart capability to permit user-directed statistical convergence. In eigenvalue calculations, the beginning source sites are from a random (flat) guess, or can be provided via ASCII input, or from a previous calculation. The starting neutrons for each subsequent generation are randomly selected from the potential fission sites in the previous generation.
Track-length or collision estimates of reaction rates are automatically tallied by energy group and edit region to facilitate comparison to other calculations. Groupwise edits include isotopic and macroscopic reaction rates and cross sections, group-to-group scattering cross sections, net currents, and scalar fluxes. Particle pseudo-collisions are used to estimate microscopic group-to-group (n,2n), inelastic, and PN elastic scattering. The serial correlation of eigenvalue estimates is computed to detect underestimated errors.