[Paper Review] Accuracy of ab initio methods in predicting the crystal structures of metals: review of 80 binary alloys

Stefano Curtarolo, Dane Morgan+1 more|arXiv (Cornell University)|Feb 18, 2005

Metallurgical and Alloy Processes5 references9 citations

TL;DR

This study evaluates the accuracy of ab initio density functional theory (DFT) using LDA/GGA functionals in predicting the ground-state crystal structures of 80 binary metallic alloys, computing 17,600 total energies across 176 crystal structures. It finds that DFT correctly predicts the experimental ground state in 89 out of 92 unambiguous cases, with only three significant discrepancies, demonstrating high predictive power for materials design.

ABSTRACT

Predicting and characterizing the crystal structure of materials is a key problem in materials research and development. We report the results of ab initio LDA/GGA computations for the following systems: AgAu, AgCd, AgMg, AgMo*, AgNa, AgNb*, AgPd, AgRh*, AgRu*, AgTc*, AgTi, AgY, AgZr, AlSc, AuCd, AuMo*, AuNb, AuPd, AuPt*, AuRh*, AuRu*, AuSc, AuTc*, AuTi, AuY, AuZr, CdMo*, CdNb*, CdPd, CdPt, CdRh, CdRu*, CdTc*, CdTi, CdY, CdZr, CrMg*, MoNb, MoPd, MoPt, MoRh, MoRu, MoTc*, MoTi, MoY*, MoZr, NbPd, NbPt, NbRh, NbRu, NbTc, NbY*, NbZr*, PdPt, PdRh*, PdRu*, PdTc, PdTi, PdY, PdZr, PtRh, PtRu, PtY, PtTc, PtTi, PtZr, RhRu, RhTc, RhTi, RhY, RhZr, RuTi, RuTc, RuY, RuZr, TcTi, TcY, TcZr, TiZr*, YZr* (*= systems in which the ab initio method predicts that no compounds are stable). A detailed comparison to experimental data confirms the high accuracy with which ab initio methods can predict ground states. Keywords: Binary Alloys, Ab initio, Intermetallics, Transition Metals, StructureAluminum, Cadmium, Gold, Magnesium, Molybdenum, Niobium, Palladium, Platinum, Rhodium, Ruthenium, Scandium, Silver, Sodium, Titanium, Technetium, Yttrium, Zirconium.

Motivation & Objective

• To evaluate the predictive accuracy of ab initio DFT (LDA/GGA) methods in determining the ground-state crystal structures of binary metallic alloys.
• To establish a comprehensive database of 176 crystal structures and 17,600 total energy calculations across 80 binary alloys for high-throughput materials screening.
• To identify and validate new, previously unobserved crystal structures predicted by ab initio calculations.
• To resolve discrepancies between experimental and ab initio predictions by assessing the reliability of experimental structure assignments.

Proposed method

• Performed all-electron ab initio total energy calculations using the LDA and GGA functionals within the projector augmented wave (PAW) and ultrasoft pseudopotential (US) approaches.
• Systematically computed the total energy for 176 distinct crystal structures across 80 binary alloys, including common prototypes like fcc, bcc, hcp, and complex intermetallics.
• Compared ab initio ground-state energies with experimental phase stability data from literature compilations (references [9,10]) to assess predictive accuracy.
• Identified new stable crystal structures via energy minimization and symmetry analysis, including superstructures of fcc, bcc, and hcp lattices, and non-standard structures.
• Used formation energy differences (ΔE_f) between experimental and ab initio predictions to quantify discrepancies, with thresholds of 10 meV/atom for near-degeneracy.
• Applied high-throughput screening to identify systems where no compounds are stable (marked with *), based on energy trends.

Experimental results

Research questions

• RQ1 How accurately can LDA/GGA-based ab initio methods predict the ground-state crystal structure of binary metallic alloys compared to experimental data?
• RQ2 What is the frequency of significant discrepancies between ab initio predictions and experimental structure assignments in binary intermetallics?
• RQ3 Which new, previously unobserved crystal structures are predicted to be stable by ab initio calculations?
• RQ4 To what extent are experimental structure assignments in the literature reliable when validated against ab initio total energy calculations?

Key findings

• Ab initio LDA/GGA calculations correctly predict the experimental ground-state structure in 89 out of 92 unambiguously assigned systems, indicating high predictive accuracy.
• Only three systems show significant discrepancies (ΔE_f > 20 meV/atom) between ab initio and experiment, suggesting that most experimental structure assignments are reliable.
• Four additional systems show near-degeneracy (ΔE_f < 10 meV/atom) between ab initio and experimental ground states, indicating potential ambiguity in experimental assignment.
• Five new crystal structures were predicted to be stable, including AB₃ superstructures on fcc, A₂B₂ and AB₃ superstructures on bcc, and A₂B₄ on hcp lattices.
• Two novel non-superstructure phases were discovered: MoZr₃ and Mo₅Ti (and related MoZr₅, Nb₅Ru), with monoclinic symmetry and complex atomic arrangements.
• For 27 compounds, the experimental structure could not be verified due to absence in the computational library, highlighting gaps in experimental characterization.