As technical support of the French Nuclear Safety Authority, IRSN has been developing the MORET Monte Carlo code for many years in the framework of criticality safety assessment (it is part of the CRISTAL French criticality package) and is now working to extend its application to reactor physics, starting from the new version 5.A.1. For that purpose, beside the validation for criticality safety (more than 2000 benchmarks from the ICSBEP database have been modeled and analyzed), a complementary validation phase for reactor physics has been started, with benchmarks from IRPHEP database and others.
In particular, to evaluate the performance of MORET for local flux or power density calculations in full size reactor cores, it has been decided to contribute to the computational benchmark proposed by Dr. Hoogenboom and Dr. Martin at the ANS M&C 2009 conference. The aim of this benchmark is to monitor, in forthcoming decades, the performance of detailed Monte Carlo full core calculations to assess the progress towards meeting the “Kord Smith challenge” (achieving, in less than one hour on a standard desk computer, less than 1 % standard deviation for the local power of each fuel pin of each assembly when subdivided in 100 axial and 10 radial zones for burnup calculations). A full PWR reactor core is modeled to compute local power densities for more than 6.3 million fuel elements.
This paper presents the first results obtained at IRSN with MCNP and MORET on this benchmark. The toughest difficulties come partly from the large number of loosely coupled fuel elements. In such a case, it has been shown that source convergence issues could cause large errors in local tallies. To avoid intra-generational correlation, biased or slow source convergence, one has to ensure enough neutrons per batch, enough batches and an initial source distribution as close to the real distribution as possible. Alternative sampling or tracking methods (such as Wielandt’s or Woodcock’s) have been implemented in MORET, and their operational effects on such a complex case have been studied. Beyond source convergence issues, with more than 6.3 million elements, it is difficult to get sufficient histories in each cell to obtain acceptable statistical uncertainties. In this particular case, IRSN obtained promising results comparing MORET and MCNP in terms of local power densities, standard deviations and computing times.