Analytic score comparison of geometry splitting and exponential transform for a simple Monte Carlo problem
The interpretation of the statistical error estimates produced by Monte Carlo transport codes is still somewhat of an art. Empirically, there are variance reduction techniques whose error estimates are almost always reliable and there are variance reduction techniques whose error estimates are often unreliable. Unreliable error estimates usually result from inadequate large score sampling from the score distribution's tail. This paper provides the analytic score distribution for geometry splitting/Russian roulette applied to a simple Monte Carlo problem and the analytic score distribution for the exponential transform applied to the same Monte Carlo problem. It is shown that the large score tails of the two distributions behave very differently. In particular, the exponential transform is shown to have an infinite variance for some parameter choices.
- Research Organization:
- Los Alamos National Lab., NM (United States)
- Sponsoring Organization:
- USDOE; USDOE, Washington, DC (United States)
- DOE Contract Number:
- W-7405-ENG-36
- OSTI ID:
- 5505519
- Report Number(s):
- LA-12258; ON: DE92013108
- Country of Publication:
- United States
- Language:
- English
Similar Records
Analytic Monte Carlo score distributions for future statistical confidence interval studies
Analytic score distributions for a spatially continuous tridirectional Monte Carol transport problem