Automatic differentiation approach for property computations in nanoscale thermal transport

摘要

We present the automatic code differentiation technique to perform derivative computations in nanoscale phonon transport simulations. This method exploits the concepts of templating and operator overloading in C++ and other similar programming languages to unintrusively convert existing codes into those yielding derivatives of arbitrary order. The idea is demonstrated through the computation of phonon properties such as second and third order force constants, the Gruneisen parameter, group velocities, and the temperature variation of specific heat for materials like graphene and graphene nanoribbons. Derivative values so computed are compared with those obtained using finite difference approaches or with analytical values. The method is found to yield derivative values to machine accuracy, with none of the round-off issues associated with finite difference approaches. (C) 2020 Elsevier B.V. All rights reserved.