Solution of large underestimation problem in the Monte Carlo calculation with hard biasing; In case with geometry input data created by CAD/MCNP automatic converter

Iida, Hiromasa; Kawasaki, Nobuo*; Konno, Chikara  ; Sato, Satoshi; Seki, Akiyuki   

An inconvenient experience was encountered, in which we have different answers depending on applied weight window values, in the nuclear analysis of the benchmark problem for CAD/MCNP interface programs, being developed under the ITER R&D task. Biasing can enhance calculation speed, but should not give different answers. Mechanism of this large underestimation is clarified. It is caused by the combination of the following two facts; (1) When one of particles in a history has got lost, MCNP cancels all tallies calculated during the history and all banked particles are thrown away (never tracked). (2) When we have distributed micro geometry errors in input data, important histories, which give significant contribution to tallies, will have many splitting and have "lost particle" with higher probability in the case of hard biasing. These two facts lead to selective canceling of important histories. An attempt to eliminate this inconvenience has been made, by modifying the subroutine "hstory" of MCNP. The modification has been done very successfully and eliminated the large underestimation, giving the same answer independently from applied weight window values.