Alchemical screening of ionic crystals

Abstract

We introduce alchemical perturbations as a rapid and accurate tool to estimate fundamental structural and energetic properties in pure and mixed ionic crystals. We investigated formation energies, lattice constants, and bulk moduli for all sixteen iso-valence-electron combinations of pure pristine alkali halides involving elements Me ∈ {Na, K, Rb, Cs} and X ∈ {F, Cl, Br, I}. For rock salt, zinc-blende, and cesium chloride symmetry, alchemical Hellmann-Feynman derivatives, evaluated along lattice scans of sixteen reference crystals, have been obtained for coupling to all respective 16 × 15 target crystals. Mean absolute errors (MAEs) are on par with the density functional theory level of accuracy for energies and bulk moduli. The predicted lattice constants are less accurate but reproduce qualitative trends. The reference salt NaCl affords the most accurate alchemical estimates of relative energies (MAE < 40 meV per atom). The best predictions of lattice constants are based on NaF as a reference salt (MAE < 0.5 Å), accounting only for qualitative trends. The best reference salt for the prediction of bulk moduli is CsCl (MAE < 0.4 × 10 dynes cm). The alchemical predictions distinguish competing rock salt and cesium chloride phases in binary and ternary solid mixtures with CsCl. Using pure RbI as a reference salt, they reproduce the reversal of the rock salt/cesium chloride stability trend for binary MeXCsCl as well as for ternary MeX(Me'Y)CsCl mixtures.

Learning physical descriptors for materials science by compressed sensing

“…It is typically considered a universal approach for learning (fitting) a complex relationship ( ) = y f x . Some, though few, machine-learning (ML) based works have been done in materials science, e.g., [2][3][4][5][6][7][8][9][10][11][12][13][14][15][16][17]. Most of them use kernel ridge regression (KRR) or Gaussian processes, where in both cases the fitted/learned property is expressed as a weighted sum over all or selected data points.…”

Alchemical Normal Modes Unify Chemical Space

“…First order terms are firmly established for all variables through the Hellmann-Feynman theorem for changes in nuclear positions (to relax or run ab initio molecular dynamics 42 ), and charges. 22,27,32,33,36,37,[43][44][45] The derivative with respect to N is related to ionization potential and electron affinity by virtue of Koopman's and Janak's theorem, 46 and exhibits the well established derivative discontinuity at integer N , 47,48 so important for the construction of improved exchangecorrelation approximations. 49 Some elements in the Hessian, molecular vibrational normal modes, or the second order derivative of the electronic energy with respect to the number of electrons is the chemical hardness, introduced by Parr and Pearson.…”

AlxGa1−xAs crystals with direct 2 eV band gaps from computational alchemy

“…For molecules and ionic crystals, even chemical accuracy can be achieved in terms of alchemical first order based estimates of relative energies when fixing number of valence electrons and geometry. 24,25 Such constraints can also be met by several material classes of great interest, such as III-V semiconductors. Here, we rely on a materials design algorithm which combines alchemical gradients with stochastic sampling and which holds promise for general computational materials design campaigns.…”