Nonlinear electron-phonon interactions from first principles
Paper details
Section titled “Paper details”| Field | Value |
|---|---|
| Journal | Physical review. B./Physical review. B |
| Published | 2026-09-15 |
| Authors | Anonymous, Feliciano Giustino |
| DOI | 10.1103/q5sl-wtlp |
Including second-order electron-phonon couplings reduced the residual error in the LiF small-hole-polaron energy curve to below 11 meV. The linear polaron equations overestimated the formation energy by as much as 0.4 eV, about 53%, and overestimated the hopping barrier by roughly a factor of two. The paper also derives a finite-difference method for complete diagonal and off-diagonal second-order coupling matrix elements.
Complete nonlinear couplings from first principles
Section titled “Complete nonlinear couplings from first principles”A second-order matrix element describes an electron changing state and wavevector through two phonons. The derived expression combines Kohn-Sham wavefunctions from the unit cell with perturbed Kohn-Sham potentials from a Born-von Karman supercell. Translating the integration variable fixes one atom in a reference cell. It reduces the required second derivatives from (3 N_at N_p)^2 to (3 N_at)^2 N_p.
Earlier solid-state approaches returned selected diagonal elements or sums rather than the full coupling matrix. The rigid-ion approximation omits mixed derivatives between different atoms. It needs 3^2 N_at supercell calculations and reduces cost by a factor of N_at N_p. For a 3 × 3 × 3 supercell with two atoms per cell, the saving is 54-fold. The same finite differences also yield first-order couplings without extra supercell calculations.
Third-order coupling follows formally from third-order finite differences. Its cost of (3 N_at)^3 N_p^2 calculations is prohibitive without an approximation. The demonstrated implementation uses norm-conserving pseudopotentials and permits semilocal or nonlocal exchange-correlation functionals.
Finite-difference workflow
Section titled “Finite-difference workflow”The unit-cell calculation first finds the self-consistent charge density. A non-self-consistent calculation then supplies Kohn-Sham states and vibrational modes on the same N1 × N2 × N3 grid. In the supercell, the authors displace one atom and compute its self-consistent distorted potential. They map unit-cell plane-wave coefficients into the supercell through G = k + g, then evaluate the matrix integral with a reciprocal-space resolution of identity.
This mapping avoids calculating unoccupied Kohn-Sham states in every supercell. It also retains unit-cell wavefunctions suitable for Wannier interpolation. Preliminary real-space results show exponential decay with electronic separation and much smaller values away from the nuclear displacement. These results support second-order Wannier interpolation.
Metals follow the proposed interpolation directly. Insulators and semiconductors require treatment of long-wavelength dipolar and quadrupolar nonanalytic terms.
Numerical validation
Section titled “Numerical validation”The calculations use Quantum ESPRESSO, LDA for diamond and graphite, and PBE for LiF. Norm-conserving pseudopotentials use a 100 Ry plane-wave cutoff. Charge-density grids are 8 × 8 × 8 for diamond, 8 × 8 × 2 for graphite, and 6 × 6 × 6 for LiF. Unless otherwise specified, coupling tests use 4 × 4 × 4 k/q grids and matching 4 × 4 × 4 supercells.
At δ = 0.01 Å, the finite-difference linear couplings agree with DFPT as follows.
| Material | δ (Å) | MAE | MARE | Bands averaged |
|---|---|---|---|---|
| Diamond | 0.01 | below 1 meV | 0.67% | 4 valence, 4 conduction |
| LiF | 0.01 | below 1 meV | 0.69% | 5 valence, 5 conduction |
| Graphite | 0.01 | below 1 meV | 0.40% | 8 valence, 8 conduction |
The second-order diamond test gives a 3 meV MAE and 0.63% MARE against the RIA Debye-Waller shift. LiF and graphite remain below 2% MARE. The remaining production calculations use δ = 0.02 Å to limit finite-difference noise. Raising the supercell energy threshold from 10^-14 Ry to 10^-16 Ry changes results by at most 0.018 meV at δ = 0.02 Å.
Rigid-ion and finite-range limits
Section titled “Rigid-ion and finite-range limits”The rigid-ion approximation is exact for independent ionic pseudopotentials moving with their ions. The paper attributes non-rigid-ion contributions to coordinate-dependent changes in the inverse electronic dielectric matrix. In diamond, 98% of full second-order elements have magnitudes below 200 meV, although some reach 2 eV. Within the ±200 meV range, full and RIA elements have an MAE of 0.67 meV and an MARE of 0.07%.
Mode-specific behavior is less uniform. For the LiF zone-center optical phonon, full elements almost exactly reproduce direct supercell valence-band-maximum shifts. The RIA overestimates those shifts by approximately a factor of two through collective lithium displacements up to 0.016 Å. At the highest transverse optical mode at X, both methods agree accurately with the supercell result.
The finite-range approximation retains selected neighbor shells.
| Material | MAE with 1st-neighbor shell | MARE | Improvement over RIA |
|---|---|---|---|
| Diamond | approximately 0.1 meV | approximately 0.01% | approximately 6× |
| LiF | approximately 0.1 meV | approximately 0.01% | approximately 3× |
Adding the second-neighbor shell gives a further factor of two in diamond and marginal changes in LiF. Although potential derivatives decay rapidly, off-diagonal q = q′ couplings can be as large as diagonal couplings. They cannot be omitted in polaron calculations.
Polaron energetics
Section titled “Polaron energetics”The test case is a small, one-center hole polaron in LiF, with atomic displacements up to 0.4 Å. The new calculations use a 3 × 3 × 3 Brillouin-zone grid and the 1st-neighbor finite-range approximation. Linear polaron equations overestimate the formation energy by as much as 0.4 eV, about 53%. They produce a potential well that is too deep and place its minimum at too large a displacement. Quadratic polaron equations reduce the residual error to below 11 meV across the tested coordinate range.
Earlier supercell and linear-equation reference values show the same overestimate.
| Reference | Quantity | Supercell DFT | Linear polaron equations |
|---|---|---|---|
| Infinite supercell | Formation energy | 1702 meV | 1980 meV |
| Infinite supercell | Excitation energy | 4176 meV | 4760 meV |
| 3 × 3 × 3 supercell | Formation energy | 652 meV | 887 meV |
| 3 × 3 × 3 supercell | Excitation energy | 2106 meV | 2487 meV |
The elastic-energy curves nearly coincide across the three methods. This agreement indicates that cubic and higher-order terms in the neutral total energy contribute little in LiF. Direct supercell calculations include them by construction, whereas the polaron equations omit them. The discrepancy instead comes from the vertical excitation energy and the neglected second-order change in the polaron eigenvalue.
Hopping and mobility
Section titled “Hopping and mobility”The hopping path moves the hole from one F atom to a nearest-neighbor F atom over R = 2.87 Å. Intermediate configurations are generated by linear interpolation between the two endpoints. The mobility estimate uses highest-energy optical modes with ℏω = 75 meV in the Emin-Holstein-Austin-Mott model.
Linear electron-phonon coupling overestimates the hopping barrier by approximately a factor of two relative to direct supercell calculations. Quadratic coupling closely matches the supercell barrier. It also recovers the correct order of magnitude of mobility, while the linear model underestimates mobility by many orders of magnitude. The authors caution that this mobility formula is an estimate and that quantitative predictions require more sophisticated approaches.
Diamond engineering implications and remaining limits
Section titled “Diamond engineering implications and remaining limits”For diamond modeling, agreement between full and RIA elements supports RIA for many summed matrix elements, including Debye-Waller corrections. Mode-specific calculations still require validation because the LiF Γ-mode test exposes an error in the RIA. The 1st-neighbor finite-range approximation provides a controlled compromise, with about 0.1 meV MAE and 0.01% MARE for diamond and LiF. Off-diagonal elements remain necessary because q = q′ couplings can reach diagonal magnitudes.
Full third-order calculation remains too expensive for the reported approach. The implementation centers on norm-conserving pseudopotentials, leaving PAW projector derivatives outside the demonstrated workflow. Wannier interpolation for insulators and semiconductors still needs a complete treatment of long-range nonanalytic terms. The reported LiF polaron calculation requires M₁ = 400 and M₂ = 70 for convergence. It also shows partial cancellation between first- and second-order contributions.
The paper does not test a device geometry, boron concentration, contact metallization, or sample dimensions. It therefore provides modeling guidance for diamond rather than criteria for selecting a doped grade or contact structure. The implementation is available in EPW under branch fdeph.
View Original Abstract
Electron-phonon interactions underpin a variety of phenomena, ranging from transport and superconductivity to polarons and ultrafast carrier dynamics. Despite being one of the most intensely studied subjects in condensed matter physics, research on electron-phonon physics mostly focused on linear, first-order couplings. Second- and higher-order nonlinear couplings are commonly ignored because their calculations are too demanding and we lack computational frameworks that can provide both diagonal and off-diagonal coupling matrix elements. In this work, we report a theory and computational method for computing nonlinear electron-phonon interactions of any order in real materials. Our approach combines the advantages of unit-cell calculations of electron wavefunctions and supercell calculations of phonon perturbations, is systematically improvable, can be used with either semilocal or nonlocal exchange-correlation functionals, and is amenable to Wannier-Fourier interpolation. As a first proof of concept, we illustrate this method by computing second-order electron-phonon coupling matrix elements in diamond, lithium fluoride, and graphite as representative nonpolar semiconductors, polar semiconductors, and metals, respectively. Furthermore, we generalize the ab initio polaron equations to second-order electron-phonon couplings, and we show that second-order couplings are essential to achieve quantitative accuracy in polaron formation energies and hopping barriers. The present methodology will find application in the study of all properties and phenomena that are currently being investigated within the linear electron-phonon coupling approximation, from phonon-mediated superconductivity to excited-states dynamics, both in harmonic and anharmonic systems.
Electronic Grade Intrinsic Diamond Epitaxial Wafer H Terminated Superior 100 Mosfet Ready Substrate
View Specs →
Boron Doped Diamond On Titanium Bdd Electrode
View Specs →