Applicability of DFT + U to U metal and U-Zr alloy

Abstract

In the Letter [J. Nucl. Mater. 444, 356 (2014)] and Comment [Phys Rev B 90, 157101 (2014)], Soderlind et al. argue that 1) DFT based on GGA already models U metal and U-Zr alloy accurately, and 2) DFT + U models them worse than DFT according to results they calculate or select from our recent study [Phys. Rev. B 88, 235128 (2013)]. Here we demonstrate in response to 1) that previously neglected and more recent experimental data indicate that DFT, even when implemented in all-electron methods, does not model the bulk modulus and elastic constants of αU very accurately. Furthermore, Soderlind et al.'s claim that deficiency exists in our PAW calculation is unfounded and hence our results, including those that show DFT results compare unfavorably with experimental/computational references, are valid. We also demonstrate in response to 2) that Soderlind et al.'s arguments are unsound for three reasons. First, they focus on just the BCC phases γU and γ(U,Zr)--which at the ab initio modeling temperature of 0 K are unstable and hence difficult to model and benchmar--as primary subjects of examination; however it is results for the stable phases mostly neglected by Soderlind et al. that are the primary evidence of our argument. Second, they make unfounded generalization of DFT + U results at selected U eff values to argue that DFT + U in general gives wrong or unsatisfactory results. Third, some key points in Soderlind et al.'s criticisms of DFT + U are not well supported, including the claims that for γ(U,Zr) DFT + U at Ueff = 1.24 eV gives positive deviations from linear dependence of composition that are unprecedentedly large and partially negative enthalpies of mixing that are inconsistent with the existence of miscibility gap in the experimental phase diagram. We therefore maintain our conclusion that DFT + U can be of value for modeling U and U-Zr.