Category:Electron-phonon interactions

In many systems it suffices to treat the electronic and vibrational degrees of freedom (phonons) separately, because electrons are much faster than the motion of nuclei. This treatment is approximate and can be corrected by including electron-phonon coupling. This entails the coupling of the two systems, while still treating the two systems separately. In fact, phonon scattering is the prevalent effect in a wide range of applications such as mobility of semiconductors or conductivity of metals at room temperature.

The inclusion of the effects of the ionic degrees of freedom in the electronic structure is important in the determination of many physical observables such as the bandgap, spectral functions, electronic conductivity, Seebeck coefficient or electronic thermal conductivity to name a few.

One can use different approaches to include electron-phonon coupling:

• statistically averaging observables during a molecular dynamics calculation,
• stochastically sampling the ionic degrees of freedom, or
• using many-body perturbation theory (available as of VASP 6.5.0).

In this page we outline the two later approaches, their advantages and disadvantages for a range of applications. For a comprehensive overview of the theory of electron-phonon coupling, we recommend the review by F. Giustino ^[1].

Stochastic displacements approach

The stochastic approach allows obtaining the bandgap renormalization and an approximation of the electronic spectral function due to the ionic degrees of freedom under the static approximation using a supercell approach. This has the advantage that it can be easily implemented and used with different levels of theory to describe the electronic states, such as different exchange-correlation functionals or even the GW approximation. The disadvantage is that the approach does not include time-dependent or dynamical effects of the phonons (static approximation) and, hence, it does not provide transport properties (see perturbative approach). Additionally, it is computationally demanding as it requires many supercell calculations.

To displace the atoms along a set of random or a single specially chosen direction ^[2], this approach requires the knowledge of the phonons on a supercell. The displacement length is determined by the temperature of the ionic system. The bandgap can be directly obtained by averaging over the set of randomly displaced supercells, or from the aforementioned special displacement pattern.

• Electron-phonon interactions from Monte-Carlo sampling

Many-body perturbation theory

Another approach to include electron-phonon coupling employs the methods and language of many-body perturbation theory, where the coupling is included as a perturbation of the electronic or phononic states. In the case of the perturbation of the electronic states, we can access the bandgap renormalization as well as electronic transport coefficients with the inclusion of phonon scattering.

This approach entails computing the electron-phonon matrix element and the phonon-induced electron self-energy. Within the framework of density-functional theory this requires the knowledge of the change of Kohn-Sham potential with an ionic perturbation as well as the initial and final electronic Kohn-Sham states. The calculation is done in two steps:

Step 1: Electron-phonon potential from supercells

In the first step, we obtain the derivatives of the Kohn-sham potential with respect to the ionic displacements

${\displaystyle \partial_{I \alpha} V (\mathbf{r}) = \frac{\partial V(r)}{\partial R_{I\alpha}} }$

with ${\displaystyle I}$ the ion index and ${\displaystyle \alpha}$ denoting the Cartesian direction in which it is displaced. The main output file is phelel_params.hdf5, which is required for computing the matrix elements in the next step.

• How to compute the electron-phonon potential from supercells

Step 2: Physical observables (or electron-phonon matrix elements)

To compute the physical observables, the phonon-induced electron self-energy is computed in the primitive cell. The main tag that provides convenient defaults depending on the observable of interest is ELPH_MODE. The computation of the self-energy requires evaluating the electron-phonon matrix elements

${\displaystyle g_{mn \mathbf{k}, \nu \mathbf{q}} \equiv \langle \psi_{m \mathbf{k} - \mathbf{q}} | \partial_{\nu \mathbf{q}} V | \psi_{n \mathbf{k}} \rangle. }$

By default, we avoid writing the matrix elements, because it is a huge data set which is distributed for optimal use of the computational resources.

For details on the setup and practical advice, we recommend reading

• Bandgap renormalization due to electron-phonon coupling
• Transport coefficients including electron-phonon scattering
• See the ELPH_DRIVER tag to obtain the electron-phonon matrix elements for further post-processing.

The standard output of the electron-phonon code is organized using so-called electron-phonon accumulators. This increases the efficiency of the code by reusing the computed electron-phonon matrix elements. For details on how to interpret the output, consult the output section on the accumulators page.

Choosing the right approach

Both the (SA) as well as the (PA) have advantages and limitations. Depending on the nature of a particular computational, there is often one approach that is much more suitable to solving it than the other. This section is dedicated to highlighting the differences and respective advantages between SA and TA so that choosing the correct approach becomes easier.

Likely the biggest deciding factor between SA and TA are the observables that can be calculated. The SA can only calculate the renormalization of the fundamental band-gap of semiconductors and insulators. In comparison, the PA can use the mathematical machinery of many-body perturbation theory to calculate a plethora of related quantities. Currently, this includes

• Band-structure renormalization, not just of band gaps
• Transport coefficients via electronic quasiparticle lifetimes

with more observables becoming available in future versions. For transport calculations, the PA is thus the only possible choice.

However, even when calculating the band-gap renormalization, there are still differences between SA and PA. For one, the PA is limited by the harmonic approximation and only considers terms in the electron-phonon interaction up to second order in the atomic displacements. These limitations do not exist in the SA, because the electron-phonon interaction is computed implicitly during the in the displaced geometry. This gives the SA a clear edge in materials that are strongly anharmonic, such as very soft materials with weakly bound atoms.

Furthermore, the SA can directly utilize higher-level exchange-correlation functionals such as and beyond-DFT methods such as the GW approximation. While it is in principle possible to integrate these features also into the PA, this is currently not supported. Therefore, when the quasiparticle shifts due to electron-electron interactions become important, it is possible to use the PA method in combination with the GW approximation ^[3].

Another key difference is how PA and SA handle polar materials. In polar materials, longitudinal optical phonons can induce long-range electrostatic fields (Fröhlich interaction) that are difficult to capture in even very large supercells. The PA can deal with this via a special treatment of the dipole interaction that explicitly accounts for the missing long-range character in finite supercells. The SA, however, has no such correction scheme, which can be detrimental for strongly polar materials. In this case, one can only try to keep increasing the supercell size in hopes of arriving at a physically meaningful result.

References