Session J10: Invited Session: How Reliable are Computed Electronic Properties?

Validation of Atomic, Molecular and Condensed Matter Calculations

The advances in calculational techniques have brought first-principles calculations to a state of frequently having their results for various physical properties agreeing will with measured results for the same properties. Validation of computational results can therefore be an important undertaking in the overall scheme of first-principles work. Along these lines, such validation does not refer to the appropriateness of approximations made in calculations which are necessary to have a practicable methodology. Rather, such validation refers to whether calculations are implemented correctly given whatever approximations one assumes. Three areas are to be discussed in turn to illustrate issues that may arise. These include density-functional total-energy calculations in atoms and ions, in which as many as four atomic-structure programs were tested (and corrected) to permit reasonably high-precision comparisons of results. Second, calculations of dielectric properties and excitation spectra of solids shall be discussed, to illustrate the myriad of choices one might be required to make in terms of approximations, and how various approximations' results can only be compared if numerical methodologies used in conjunction with the respective approximations are sufficiently similar. Third, as a point of reference, brief consideration shall be given to the progress in the calculational validation area made within the quantum chemistry community.   

New Tools for the Verification and Validation of Electronic Structure Calculations

First-principles electronic structure calculations are playing an increasingly important role in the prediction of materials properties and in the interpretation of experimental data. Numerous simulation codes including various levels of approximations and various numerical approaches are now readily available to users. The complexity of first-principles calculations and the large number of input parameters needed in a simulation make it challenging to obtain high quality, reproducible data. Results obtained with different electronic structure codes are difficult to compare as they involve a multitude of data formats, making the process of verification and validation (V\&V) of electronic structure data complex and error prone. In order to facilitate V\&V activities, we have introduced ESTEST [1,2], a web-based framework that allows for automatic comparison and post-processing of results obtained with six electronic structure codes. Recent developments make it possible to extend this functionality to a decentralized network of servers. We discuss general issues related to the process of verification and validation of electronic structure data and outline requirements for the development of future V\&V tools. \\[4pt] [1] G. Yuan and F. Gygi, Computational Science \& Discovery 3, 015004 (2010) doi:10.1088/1749-4699/3/1/015004G\\[0pt] [2] http://estest.ucdavis.edu   

Single-electron excitations of molecules and solids -- reliability and robustness of density-functional theory and \textit{GW} calculations

State-of-the-art theory addresses single-electron excitations in molecules and condensed matter by linking density-functional theory (DFT) with many-body perturbation theory. Experimentally such results correspond to measurements by direct or indirect photoemission. In actual calculations it is common to employ the pseudopotential approach, where pseudo-wave-functions enter the calculation of the selfenergy, and the core-valence interaction is treated at the DFT level. Furthermore, calculations are typically not done selfconsistently but as a first-order perturbation on some starting point. The latter may be DFT with LDA, GGA, LDA+$U$, HF, or hybrid functionals. Unfortunately, these different starting points can give noticeably different results. In this talk I will evaluate the various approximations, by comparing and analyzing pseudopotential and all-electron calculations. I will also emphasize the need for selfconsistency either by an iterative solution of the Dyson equation or by properly adjusting the zero-th order exchange-correlation functional.   

Role of Validation and Predictions in Modeling: Specific Examples from Semiconductor Industry Applications

With the advent of newer non-Silicon materials, using modeling to estimate properties are becoming necessary for process technology development. Since these materials are integrated as part of larger devices, interfaces and material domains are increasingly modulating properties of materials. Unlike in bulk materials, electronic and thermodynamic properties are difficult to characterize in these material structures as the device sizes overlap with material domains. We will illustrate specific cases from semiconductor processing and property estimation on the importance of verification of models for internal consistency and validation with experimental data. Given the discrepancy of scales between predictions and measurement, techniques need to bridge them. In addition, models that are developed need to be modular with open interfaces for cross-checking and integration across scales as indicated in the recently announced Materials Genome Initiative.\\[4pt] [1] President's initiative on Materials Genome Initiative for Global Competitiveness, June 2011\\[2pt] [2] S. Shankar, B. V. McKoy, W. L. Morgan, ``Self-Consistent Modeling of Weakly Ionized Plasmas-Challenges in Quantum and Classical mechanics,'' Sixth U.S. National Congress on Computational Mechanics, U.S. Association for Computational Mechanics, Dearborn, Michigan, (2001)