MCNP implementation greatly increases runtime

[jbae11]

Investigated and organized by Katarzyna Borowiec at ORNL (@KatarzynaBorowiec)

Problem description

Implementation of the user defined neutron source caused significant slowdown of MCNP run. The following slowdown was observed (from 3.5M histories/hr using SDEF to 0.02M histories/hr for user-defined source). Upon investigation, this is caused by source particles being tracked until the statistical weight of the particle reaches 0.0 for the user defined source.

Identified cause

The default value of the swtm variable is set to 0. The variable is used in the MCNP initialization to determine the weight cutoffs. If the user defined source is used, swtm remains equal to 0 and the weight cutoffs are also set to 0. If the general source (e.g., SDEF card) is used this variable is set to the minimum weight of source particle.

Possible solutions

1. The swtm variable can be specified using the CUT card in the input file. It would be helpful to explicitly state the need for this card in the repo's manual or readme.
2. Changes to the MCNP source code changing the default of the swtm at the initialization stage.
3. Explicitly evaluating/setting weight cutoffs in the user defined subroutine.

[makeclean]

Thanks for reporting, do you have a preferred solution?