Nuclear Science and Engineering / Volume 193 / Number 7 / July 2019 / Pages 681-721
Technical Paper / dx.doi.org/10.1080/00295639.2018.1564504
For over 60 years, the Roussopoulos and Schwinger functionals have been used in many works and textbooks under the assumption that they provide “second-order accurate” trial functions for the forward and adjoint fluxes when computing reaction rates and/or particle detector responses in source-driven nuclear systems. The Schwinger functional has been employed as a particularly useful form of the Roussopoulos functional for systems in which the forward and adjoint particle fluxes were normalized. When using these functionals, however, the expressions for the approximate fluxes were postulated arbitrarily while the system parameters were unrealistically assumed to be perfectly well known. This work revisits the Roussopoulos and Schwinger functionals within the realistic practical context of imprecisely known model parameters, including imprecisely known cross sections, number densities, fission spectra, and forward and adjoint sources. By applying the Second-Order Adjoint Sensitivity Analysis (2nd-ASAM) methodology, this work shows that the first-order sensitivities of the Roussopoulos and Schwinger functionals to model parameters are not identically zero. This fact implies that neither the Roussopoulos nor the Schwinger functionals are accurate to second order in parameter variations/uncertainties, which implies, in turn, that these functionals are not accurate to second order variations in the flux when such flux-variations are caused by imprecisely known model parameters. Furthermore, the 2nd-ASAM methodology applied in this work also provides exactly and efficiently all of the second-order sensitivities of the Roussopoulos and Schwinger functionals to the imprecisely known model parameters. The new results presented in this work place in the correct light the results published hitherto in works that have used the Roussopoulos and Schwinger functionals while also indicating the correct path for future possible uses of these functionals for performing sensitivity and uncertainty analyses of both forward and inverse problems in nuclear systems.