Exploring A Two-Level, Control-Variate Monte Carlo Method for Uncertainty Quantification in Criticality Safety

Presented is an application of a two-level, control-variate Monte Carlo (CVMC) scheme to accelerate uncertainty quantification in criticality safety applications. The current study used data from past work on the biases introduced in K_eff due to uncertainties in the composition packaged nuclear fuel. The original analysis employed SCALE-6.2.4 and its Sampler sequence to propagate fuel composition uncertainties for each of several cases of interest. Each case required 1,000, ≈10-hr-long KENO simulations that used independent samples of the uncertain composition and produced estimates with standard errors of 20-30 pcm. By using 50 of those samples to construct an approximate control variate based on a low-rank, linear model, the CVMC produced estimates with similar standard using as few as 22 simulations (using a linear model of rank 30) and as much as 328 simulations errors (using a linear model of rank 10). These results and the simplicity of the method suggest its potential utility for similar applications.