Comparison of algorithms for Doppler broadening pointwise tabulated cross sections

In this paper, three algorithms for Doppler broadening pointwise tabulated cross sections are compared on the basis of speed and accuracy. The use of a Gauss-Legendre quadrature for numerical integration on subintervals of the convolution integral for which the cross section is smooth was found to reduce run time by 48% compared to the exact method used in NJOY and other cross section processing codes. However, a comparable decrease in run time was also achieved simply by using a more efficient implementation of the erfc function. The use of a Gauss-Hermite quadrature for performing the convolution integral resulted in an order of magnitude decrease in run time for broadening. The accuracy of the Gauss-Hermite quadrature was also studied, and it was found that quadrature orders greater than 15 reach an asymptotic level of error. Lastly, an on-the-fly implementation of Doppler broadening in OpenMC using a Gauss-Hermite quadrature resulted in at least a factor of 15 increase in run time relative to a simulation without Doppler broadening for the BEAVRS benchmark problem, underscoring the need for advanced methods for on-the-fly temperature treatment for Monte Carlo codes.