We present large-scale simulations of photovoltaics materials in the Vienna Ab initio Simulation Package [1] (VASP) that are only possible by pushing the boundaries of practically solvable eigenproblems. Thus, we enable analyses for semiconductors that were inaccessible within state-of-the-art predictive methods before. To achieve this, we a) implemented a distributed eigensolver in cuSOLVERMp that fully runs on the GPU, b) adapted VASP's Bethe-Salpeter Equation (BSE) algorithm for GPUs and to employ cuSOLVERMp, and c) dramatically improved the BSE workload distribution to yield near perfect load-balancing at scale. With a size of two million, we solve one of the largest dense, complex eigenproblems reported so far in under 5 hours sustaining 7.8 PFLOPS using 34.5 GJ on 4096 GPUs of NVIDIA's Selene supercomputer. Our results facilitate new breakthroughs in material science. These improvements in compute and energy efficiency apply to other domains relying on solving large eigenproblems, as well.

We present results of a porting effort of VASP (the Vienna Ab Initio Simulation Package) to GPUs, using OpenACC. While having been useful to researchers, the existing CUDA C based port of VASP was hard to maintain due to source code duplication. We demonstrate a directive based OpenACC adaptation for the most important DFT-level solvers available in VASP: RMM-DIIS and blocked-Davidson. A comparative performance study shows that the OpenACC efforts can even significantly outperform the former port. No extensive code refactoring was necessary. Guidelines to managing device memory for heavily aggregated data structures are presented. These lead to cleaner code and lower the entry barrier to accelerate additional parts of VASP and might prove useful for accelerating other high-performance

A hexagonal phase in the ternary Ge-Se-Te system with an approximate composition of GeSe0.75 Te0.25 has been known since the 1960s but its structure has remained unknown. We have succeeded in growing single crystals by chemical transport as a prerequisite to solve and refine the Ge4 Se3 Te structure. It consists of layers that are held together by van der Waals type weak chalcogenide-chalcogenide interactions but also display unexpected Ge-Ge contacts, as confirmed by electron microscopy analysis. The nature of the electronic structure of Ge4 Se3 Te was characterized by chemical bonding analysis, in particular by the newly introduced density of energy (DOE) function. The Ge-Ge bonding interactions serve to hold electrons that would otherwise go into antibonding Ge-Te contacts.

AbstractIm ternären System Ge‐Se‐Te ist seit den 1960er Jahren eine hexagonale Phase mit der annähernden Zusammensetzung GeSe0.75Te0.25 bekannt, aber ihre Struktur blieb ungeklärt. Durch chemischen Transport gelang nun die Zucht von Einkristallen zur Bestimmung der Ge4Se3Te‐Struktur. Diese besteht aus Schichten, die – wie elektronenmikroskopische Analysen erhärten – über van‐der‐Waals‐artige schwache Chalkogen‐Chalkogen‐, aber auch über unerwartete Ge‐Ge‐Wechselwirkungen verbunden sind. Die Art der elektronischen Struktur von Ge4Se3Te wurde durch Analyse der chemischen Bindungen bestimmt, speziell mithilfe der neu eingeführten Energiedichte(DOE)‐Funktion. Die bindenden Ge‐Ge‐Wechselwirkungen dienen der Aufnahme von Elektronen, die ansonsten antibindende Ge‐Te‐Zustände einnähmen.

Plotting materials on bi‐coordinate maps according to physically meaningful descriptors has a successful tradition in computational solid‐state science spanning more than four decades. Equipped with new ab initio techniques introduced in this work, we generate an improved version of the treasure map for phase‐change materials (PCMs) as introduced previously by Lencer et al. which, other than before, charts all industrially used PCMs correctly. Furthermore, we suggest seven new PCM candidates, namely SiSb4Te7, Si2Sb2Te5, SiAs2Te4, PbAs2Te4, SiSb2Te4, Sn2As2Te5, and PbAs4Te7, to be used as synthetic targets. To realize aforementioned maps based on orbital mixing (or “hybridization”) and ionicity coordinates, structural information was first included into an ab initio numerical descriptor for sp3 orbital mixing and then generalized beyond high‐symmetry structures. In addition, a simple, yet powerful quantum‐mechanical ionization measure also including structural information was introduced. Taken together, these tools allow for (automatically) generating materials maps solely relying on first‐principles calculations. © 2017 Wiley Periodicals, Inc.

Unerwartete Ge-Ge-Wechselwirkungen im “zweidimensionalen” Ge4Se3Te fanden R. Dronskowski et al., wie in der Zuschrift auf S. 10338 berichtet wird. Die Schichtverbindung wurde mittels chemischen Transports kristallisiert und erstmals strukturell charakterisiert. Ihre elektronische Struktur und der Ursprung der Ge-Ge-Wechselwirkungen wurden mithilfe chemischer Bindungsanalyse und der neu eingeführten Energiedichte(DOE)-Funktion beleuchtet.

Unexpected Ge–Ge interactions were found in “two-dimensional” Ge4Se3Te as reported by R. Dronskowski et al. in their Communication on page 10204 ff. The layered material was crystallized using chemical vapor deposition and then characterized for the first time. Its electronic structure and the chemical cause of the Ge–Ge interactions were examined by chemical bonding analysis and the newly introduced density of energy (DOE) function.

A well-defined spatial orientation of atomic basis functions is essential for the correct analysis of quantum-mechanical calculations in terms of chemical (bonding) concepts. Here, we present the implementation of a straightforward, convenient algorithm to rotate basis functions using real spherical harmonics within a linear combination of atomic orbitals (LCAO) framework. The highly efficient technique only relies on overlap integrals of the basis functions and Wigner's rotation matrices. To do so, a previously known and simple way to calculate the latter (defined by a rotation axis and angle) for real spherical harmonics is modified to enable chemical-bonding interpretation. The method's usefulness is illustrated by an application to carbon crystallizing in the diamond structure.

The computer program LOBSTER (Local Orbital Basis Suite Towards Electronic‐Structure Reconstruction) enables chemical‐bonding analysis based on periodic plane‐wave (PAW) density‐functional theory (DFT) output and is applicable to a wide range of first‐principles simulations in solid‐state and materials chemistry. LOBSTER incorporates analytic projection routines described previously in this very journal [J. Comput. Chem. 2013 , 34 , 2557] and offers improved functionality. It calculates, among others, atom‐projected densities of states (pDOS), projected crystal orbital Hamilton population (pCOHP) curves, and the recently introduced bond‐weighted distribution function (BWDF). The software is offered free‐of‐charge for non‐commercial research. © 2016 The Authors. Journal of Computational Chemistry Published by Wiley Periodicals, Inc.

Abstractα‐CuN3 precipitates from aqueous solutions of CuCl2, Na2SO3, and NaN3.

Despite its simple chemical constitution and unparalleled technological importance, the phase-change material germanium telluride (GeTe) still poses fundamental questions. In particular, the bonding mechanisms in amorphous GeTe have remained elusive to date, owing to the lack of suitable bond-analysis tools. Herein, we introduce a bonding indicator for amorphous structures, dubbed "bond-weighted distribution function" (BWDF), and we apply this method to amorphous GeTe. The results underline a peculiar role of homopolar Ge-Ge bonds, which locally stabilize tetrahedral fragments but not the global network. This atom-resolved (i.e., chemical) perspective has implications for the stability of amorphous "zero bits" and thus for the technologically relevant resistance-drift phenomenon.

An unexpected polymorph of the highly energetic phase CuN3 has been synthesized and crystallizes in the orthorhombic space group Cmcm with a=3.3635(7), b=10.669(2), c=5.5547(11) Å and V=199.34(7) Å(3). The layered structure resembles graphite with an interlayer distance of 2.777(1) Å (=1/2 c). Within a single layer, considering N3(-) as one structural unit, there are 10-membered almost hexagonal rings with a heterographene-like motif. Copper and nitrogen atoms are covalently bonded with Cu-N bonds lengths of 1.91 and 2.00 Å, and the N3(-) group is linear but with N-N 1.14 and 1.20 Å. Electronic-structure calculations and experimental thermochemistry show that the new polymorph termed β-CuN3 is more stable than the established α-CuN3 phase. Also, β-CuN3 is dynamically, and thus thermochemically, metastable according to the calculated phonon density of states. In addition, β-CuN3 exhibits negative thermal expansion within the graphene-like layer.

A glimpse into amorphous materials. In their Communication on page 10817 ff. R. Mazzarello, R. Dronskowski et al. investigate the bonding nature of amorphous GeTe, a prototypical phase-change data-storage material. Starting from a molecular-dynamics snapshot, the role of different structural motifs is elucidated by the use of new tools; this way, the chemical “language” of bonding and coordination polyhedra may be transferred to the amorphous state.

AbstractUngeachtet seiner einfachen Zusammensetzung und enormen technischen Wichtigkeit wirft das Phasenwechselmaterial Germaniumtellurid (GeTe) auch heute noch grundlegende Fragen auf. Insbesondere die Bindungsmechanismen in amorphem GeTe haben sich bislang einer schlüssigen Erklärung entzogen, da geeignete bindungsanalytische Hilfsmittel fehlten. Hier führen wir einen neuen Bindungsindikator für amorphe Strukturen ein, den wir als “bindungsgewichtete Verteilungsfunktion” (BWDF) bezeichnen, und wenden diese Methode auf amorphes GeTe an. Die Ergebnisse belegen, daß homopolare Ge‐Ge‐Bindungen darin eine besondere Rolle spielen: sie stabilisieren einzelne tetraedrische Fragmente, nicht aber das globale Netzwerk. Diese atomaufgelöste (d. h. chemische) Perspektive hat Bedeutung für die Langzeitstabilität amorpher “Null‐Bits” und damit für das technisch wichtige Phänomen der Widerstandsdrift.

AbstractWir berichten über die Synthese einer unerwarteten Modifikation der hochgradig energiereichen Phase CuN3, die in der orthorhombischen Raumgruppe Cmcm mit a=3.3635(7), b=10.669(2), c=5.5547(11) Å und V=199.34(7) Å3 kristallisiert. Die schichtartige Struktur ähnelt Graphit mit einem Schichtabstand von 2.777(1) Å (=1/2 c). Sofern man N3− als strukturelle Einheit ansieht, liegen innerhalb einer Schicht annähernd hexagonale Zehnerringe mit heterographenartigem Motiv vor. Kupfer‐ und Stickstoffatome sind mit Cu‐N=1×1.91 und 2×2.00 Å kovalent aneinander gebunden, und die N3−‐Gruppe fällt mit N‐N=1.14 und 1.20 Å linear aus. Berechnungen der Elektronenstrukturen und experimentelle Thermochemie ergeben, daß die als β‐CuN3 bezeichnete neue Modifikation stabiler als die bekannte Phase α‐CuN3 ist. Des weiteren erweist sich β‐CuN3 anhand der berechneten phononischen Zustandsdichte als dynamisch und deshalb auch thermochemisch metastabil. β‐CuN3 zeigt in den heterographenartigen Schichten eine negative thermische Ausdehnung.

Quantum-chemical computations of solids benefit enormously from numerically efficient plane-wave (PW) basis sets, and together with the projector augmented-wave (PAW) method, the latter have risen to one of the predominant standards in computational solid-state sciences. Despite their advantages, plane waves lack local information, which makes the interpretation of local densities-of-states (DOS) difficult and precludes the direct use of atom-resolved chemical bonding indicators such as the crystal orbital overlap population (COOP) and the crystal orbital Hamilton population (COHP) techniques. Recently, a number of methods have been proposed to overcome this fundamental issue, built around the concept of basis-set projection onto a local auxiliary basis. In this work, we propose a novel computational technique toward this goal by transferring the PW/PAW wavefunctions to a properly chosen local basis using analytically derived expressions. In particular, we describe a general approach to project both PW and PAW eigenstates onto given custom orbitals, which we then exemplify at the hand of contracted multiple-ζ Slater-type orbitals. The validity of the method presented here is illustrated by applications to chemical textbook examples-diamond, gallium arsenide, the transition-metal titanium-as well as nanoscale allotropes of carbon: a nanotube and the C60 fullerene. Remarkably, the analytical approach not only recovers the total and projected electronic DOS with a high degree of confidence, but it also yields a realistic chemical-bonding picture in the framework of the projected COHP method.

We report about a source-code modification of the density-functional program suite VASP which greatly benefits from the use of graphics-processing units (GPUs). The blocked Davidson iteration scheme (EDDAV) has been optimized for GPUs and gains speed-ups of up to 3.39 on S1070 devices and of 6.97 on a C2050 device. Using the Fermi card, the code reaches an impressive 61.7% efficiency but does not suffer from any accuracy losses. The algorithmic bottleneck lies in the multiplication of rectangular matrices. We also give some initial thoughts about introducing a different level of parallelism in order to harness the computational power of multi-GPU installations.

We report on a source-code modification of the density-functional program suite VASP which benefits from the use of graphics-processing units (GPUs). For the electronic minimization needed to achieve the ground state using an implementation of the blocked Davidson iteration scheme (EDDAV), speed-ups of up to 3.39 on S1070 devices or 6.97 on a C2050 device were observed when calculating an ion–conductor system of actual research interest. Concerning the GPU specialty – memory throughput – the low double-precision performance forms the bottleneck on the S1070, whereas on Fermi cards the code reaches 61.7% efficiency while not suffering from any accuracy losses compared to well-established calculations performed on a central processing unit (CPU). The algorithmic bottleneck was found to be the multiplication of rectangular matrices. An initial idea to solve this problem is given.