Use of deterministic codes for ''separating the wheat from the chaff'' in benchmark models and calculations

摘要

An important effort in the field of nuclear criticality safety is the establishing and modeling of a set of benchmark critical (or near critical) experiments and the use of these benchmark experiment models to validate and verify computer codes and cross sections. For the most part Monte Carlo codes such as MCNP and KENO.VA have emerged as the codes of choice in the U.S. Since Monte Carlo codes generally have the capability of modeling complex geometries in great detail there is a tendency to focus attention on modeling an experiment in very great detail. Indeed, it is the author's observation that so much effort is expended trying to exactly model unimportant details (the 'chaff') that many of the truly significant features and characteristics (the 'wheat') of the experiment can be lost in the shuffle. An obvious question now arises. How does one determine what is 'wheat' and what is 'chaff'. This presentation will focus on the use of deterministic, or, more specifically, discrete-ordinates codes for performing sensitivity calculations in determining which details are important and which are not. The principal advantage of discrete-ordinates codes is that they can quite precisely determine the effect on k(sub eff) associated with distinct details of an experiment such as impurities, external structure, conflicting or missing information, etc. Three different examples are used to demonstrate the value of using discrete-ordinates codes for separating the wheat from the chaff. It is also shown that, for all practical purposes, Monte Carlo codes are essentially unsuited for determining small effects on k(sub eff).