Efficientk⋅pmethod for the calculation of total energy and electronic density of states

Marcella Iannuzzi, Michele ParrinelloPhysical 2001 · DOI

TL;DR

This paper introduces an efficient computational method for calculating the electronic structure of large periodic systems. By extending density-functional perturbation theory with a $\mathbf{k}\cdot\mathbf{p}$ approximation, it significantly reduces the computational cost of accurately sampling the Brillouin zone, enabling more reliable electronic density of states calculations for large systems.

What Problem Does This Paper Address?

Calculating the electronic structure of materials, particularly in large systems relevant to molecular dynamics simulations, typically requires sampling many points in the Brillouin Zone (BZ). This exhaustive sampling is computationally prohibitive for large systems, often forcing researchers to rely on approximations like only sampling the $\Gamma$ point. While this is acceptable for very large systems or insulators with wide band gaps, it leads to inaccurate electronic densities of states (DOS) and can miss crucial electronic behavior in many materials. Previous attempts to address this using $\mathbf{k}\cdot\mathbf{p}$ methods have been limited by computational cost or restrictions to simpler approximations. This paper aims to provide a computationally efficient and accurate method for BZ sampling in large systems, improving the calculation of total energies and electronic DOS.

Methods

The core of the method is an adaptation of density-functional perturbation theory (DFPT) to incorporate $\mathbf{k}\cdot\mathbf{p}$ corrections. The approach starts with a $\Gamma$-only calculation (sampling only the $\Gamma$ point of the BZ) to obtain the ground-state wavefunctions. Then, using DFPT, it calculates the second-order corrections to the energy as a function of $\mathbf{k}$. This involves determining first-order corrections to the wavefunctions ($\mathbf{u}_{\alpha m}^{(1)}$) and then using these to evaluate the energy dispersion. The method leverages the fact that for semiconductors and insulators, the electronic charge density is primarily determined by occupied states, and the $\mathbf{k}\cdot\mathbf{p}$ expansion can be truncated at second order in $\mathbf{k}$ for accurate energy calculations. The authors also detail how to handle non-local pseudopotentials within this framework by expanding their reciprocal space representation.

Datasets

The paper evaluates the method on two well-studied crystalline systems:

  • Argon (Ar) supercells: Used to demonstrate the rapid convergence of the method for systems with flat electronic bands.
  • Silicon (Si) supercells: Used to showcase the method's effectiveness for a prototypical semiconductor with broader electronic bands, highlighting significant improvements in energy accuracy and DOS calculations compared to $\Gamma$-only approximations.

The calculations were performed using standard pseudopotential approximations (Goedecker for Ar, Troullier-Martins for Si) and density functional approximations (LDA).

Key Insights and Ideas

  • Bridging the gap between $\Gamma$-only and full BZ sampling: The $\mathbf{k}\cdot\mathbf{p}$ DFPT approach offers a computationally feasible way to achieve results comparable to full BZ sampling, even for moderately sized supercells.
  • Accurate DOS for large systems: The method significantly improves the description of the electronic density of states, which is a critical quantity for understanding material properties and is poorly represented by $\Gamma$-only calculations.
  • Efficiency gains: The computational cost is reduced by orders of magnitude compared to full BZ sampling, making ab-initio molecular dynamics simulations with converged BZ sampling feasible for larger systems.
  • Handling non-local pseudopotentials: The paper demonstrates a practical way to incorporate the effects of non-local pseudopotentials within the $\mathbf{k}\cdot\mathbf{p}$ expansion, a crucial aspect for realistic material simulations.

Key Results

  • For Argon supercells, the $\mathbf{k}\cdot\mathbf{p}$ approach yields results practically indistinguishable from full BZ sampling even for a 32-atom supercell.
  • For Silicon, the DFPT correction provides a significant improvement in total energy accuracy. For a 128-atom supercell, the energy error is reduced by an order of magnitude compared to $\Gamma$-only calculations, and for a 250-atom supercell, the error is less than 0.001%.
  • The calculated electronic density of states for Si supercells using the $\mathbf{k}\cdot\mathbf{p}$ DFPT method closely matches the results obtained from a standard, computationally intensive full BZ calculation.
  • The method allows for the calculation of electronic DOS for a 128-atom Si supercell with 4096 k-points on a standard Intel Xeon PC, a task that would be prohibitively expensive with traditional methods.

Limitations and Future Work

  • System size dependence: While the method is efficient for large systems, the convergence rate for the $\mathbf{k}\cdot\mathbf{p}$ expansion depends on the material's band structure. Broader bands (like in Si) require more careful treatment than flatter bands (like in Ar).
  • Focus on semiconductors/insulators: The current formulation is restricted to doubly occupied states in semiconductors and insulators. Extending it to metals, where Fermi surface integration is more complex, would be a valuable future direction.
  • Further exploration of non-local PP expansion: While the paper shows the expansion of non-local PP projectors, further investigation into the accuracy and limitations of this expansion could be beneficial.
  • Application to molecular dynamics: The authors suggest that this method makes efficient ab-initio molecular dynamics calculations with fully converged BZ sampling possible, implying future work could focus on applying it to complex dynamic processes.

Summary updated

Abstract

An efficient method for calculating the electronic structure in large systems with a fully converged Brillouin zone sampling is presented. The method is based on a $\mathbf{k}\ensuremath{\cdot}\mathbf{p}$-like approximation developed in the framework of the density-functional perturbation theory. The reliability and efficiency of the method are demonstrated in test calculations on Ar and Si supercells.
BibTeX
@inproceedings{iannuzzi2001efficient,
  author = {Marcella Iannuzzi and Michele Parrinello},
  title = {{Efficient<b>k</b>⋅<b>p</b>method for the calculation of total energy and electronic density of states}},
  booktitle = {Physical},
  year = {2001},
  doi = {10.1103/physrevb.64.233104},
}
Figure