Benchmarking the Elastic Tensors of Silicon Polymorphs Across Three Density Functional Theory Databases Disconfirms a Predicted Auxetic Candidate
Abstract
We benchmark the structural and elastic properties of silicon polymorphs (diamond-Si, BC8-Si, R8-Si, Ibam-Si, and phases identified through high-throughput screening) across three independent computational databases: Materials Project, AFLOW, and JARVIS-DFT. Calibrating against the one phase with a complete experimental elastic tensor (diamond-Si), all three databases show a consistent 9-10% systematic underestimate of the bulk modulus relative to experiment, quantifying a known GGA-PBE limitation across three unrelated codes and workflows. We retrieve and cross-validate the complete elastic tensors of BC8-Si and R8-Si for the first time in a consolidated, source-traceable dataset, confirming both phases mechanically stable and finding BC8-Si to be the most elastically isotropic entry among fifteen independently characterized silicon structures on a symmetry-independent anisotropy index. An apparent auxetic silicon candidate identified by high-throughput screening (derived Poisson ratio -0.181) is shown, via a direct Born-stability-criterion violation in its full elastic tensor, to be mechanically unstable rather than a genuine effect -- illustrating a general pitfall of screening on isotropic averages alone. A component-by-component comparison of the full elastic tensor across eighteen matched AFLOW-JARVIS structures reveals that this agreement is not uniform: two of six diagonal constants, C44 and C66, show median disagreements of 13% and 19% respectively, an order of magnitude larger than the 2-4% typical of normal-stress and off-diagonal constants. This disagreement is absent in cubic structures and present in every non-cubic structure examined; re-pairing C44 against C66 collapses the discrepancy to a few percent in four independent cases, consistent with an axis-labelling convention difference between the two databases rather than a physical one, though genuine structural non-equivalence between matched entries cannot be fully excluded. Mapping the elastic-property space of silicon polymorphs in the style of Ashby materials-selection charts, we find that structural compactness predicts bulk stiffness but has no detectable power to predict ductile-versus-brittle character, a decoupling replicated independently across two of the three databases. We document a genuine, unresolved discrepancy of 11-12% between databases in the bulk modulus of β-Sn-Si, illustrating that cross-database benchmarking remains necessary even for a well-studied elemental system. All underlying data, including full elastic tensors, structure identifiers and evidence-level annotations, are made available in a source-traceable database.
Abstract as supplied by the indexing source. © the authors / publisher.