Bulk and surface energetics of crystalline lithium hydride: Benchmarks from quantum Monte Carlo and quantum chemistry

Abstract

We show how accurate benchmark values of the surface formation energy of crystalline lithium hydride can be computed by the complementary techniques of quantum Monte Carlo (QMC) and wavefunctionbased molecular quantum chemistry. To demonstrate the high accuracy of the QMC techniques, we present a detailed study of the energetics of the bulk LiH crystal, using both pseudopotential and all-electron approaches. We show that the equilibrium lattice parameter agrees with experiment to within 0.03 %, which is around the experimental uncertainty, and the cohesive energy agrees to within around 10 meV per formula unit. QMC in periodic slab geometry is used to compute the formation energy of the LiH (001) surface, and we show that the value can be accurately converged with respect to slab thickness and other technical parameters. The quantum chemistry calculations build on the recently developed hierarchical scheme for computing the correlation energy of a crystal to high precision. We show that the hierarchical scheme allows the accurate calculation of the surface formation energy, and we present results that are well converged with respect to basis set and with respect to the level of correlation treatment. The QMC and hierarchical results for the surface formation energy agree to within about 1 %.

Accurate surface energies from first principles

“…The CC and QMC values were calculated using all-electron approaches, i.e., by treating the core electrons in the same way as the valence electrons. The all-electron results indicated that core-valence correlations contributed appreciably to the overall surface energy of LiH, accounting for approximately 0.03 J/m 2 according to CC calculations [18]. However, because such calculations rely on -point-only sampling, large repeated supercells are needed.…”

“…The data for LiH are used as a benchmark because Binnie and coworkers [18] have calculated the surface energy of this material using two high-level methods: hierarchical coupled cluster (CC) [34] and quantum Monte Carlo (QMC) [35]. These methods predict surface energies of 0.434 J/m 2 and 0.44(1) J/m 2 , respectively, for LiH(001).…”

The role of density functional theory methods in the prediction of nanostructured gas-adsorbent materials

“…For these reasons, we envisage that QMC methods will play an increasingly more relevant role in prospective DFT benchmark studies of GAM. (Actually, the DMC method has already been applied to the study of metal hydrides [228,229,287] and metal oxides [288][289][290], two important families of materials within the context of hydrogen storage and carbon capture.) It is worth noticing, however, that QMC methods are neither exempt of some important technical problems.…”

“…These error cancellations, however, do not occur systematically and may depend on the specific details of GAM (e.g., see work [227] where the sign of the difference between the GGA and vdW-DF2 energies varies with the dopant species), and thus the use of standard DFT methods for modeling of carbon-based H 2 -storage materials is not recommended. Meanwhile, new quantum Monte Carlo (QMC) and RPA-DFT simulations, this time performed in periodic systems, are highly desirable for rigorously evaluating the performance of meta, hybrid, and dispersion DFT functionals (see works [152,195,[228][229][230] for examples of applications of such advanced computational methods to simulation of relevant materials). QMC and RPA-DFT calculations are also necessary for determining the relevance of many-body energy and Coulomb screening ef-fects on the present class of GAM, which so far have been systematically neglected.…”

Explicitly correlated plane waves: Accelerating convergence in periodic wavefunction expansions

“…Nowhere is this more true than in the solid state, where application of high-level quantum chemistry methods are only beginning to emerge in a recently growing field [2][3][4][5][6][7][8][9][10][11][12][13][14][15][16][17]. The reason for this slow uptake is the computational cost of these methods, which generally scale as a high power of the system size, compared to the lower mean-field scaling of DFT.…”