Skip to content

First-Principles Predictions of Carrier Mobility with Record Accuracy Using GW Perturbation Theory

FieldValue
JournalPhysical Review Letters
Published2026-05-15
AuthorsAnonymous, Sabyasachi Tiwari, Steven G. Louie, Zhenglu Li, Feliciano Giustino
DOI10.1103/83vn-7zyw

The mean absolute relative error in predicted electron mobility fell to 11% for five benchmark semiconductors at 300 K. This result used GW bands with GW perturbation theory electron-phonon couplings. The benchmark covered single-crystal Si, GaAs, GaP, diamond, and 3C-SiC. The authors report record accuracy for fully ab initio mobility calculations.

Phonon motion scatters electrons and sets an intrinsic limit on carrier mobility. The scattering rate follows the phonon-induced Fan-Migdal self-energy. Standard DFT and DFPT calculations replace the many-body dielectric response with the DFT dielectric response. Semi-local functionals can then overscreen the electron-phonon vertex. This underestimates coupling strength and reduces the predicted scattering rate.

GW band corrections can improve dispersions and effective masses, but they do not repair a DFPT electron-phonon vertex. The hybrid GW+DFPT scheme therefore gave worse agreement than DFT+DFPT for diamond, 3C-SiC, and GaP. For these materials, larger coupling corrections lowered mobility despite slightly reduced effective mass.

Mobility scales with carrier lifetime and inversely with effective mass. Both quantities therefore depend on the quality of the band and coupling calculations. The authors call the fully many-body calculation GWBTE.

The authors combined Quantum ESPRESSO, ABINIT, BerkeleyGW, Wannier90, and EPW. DFT and DFPT supplied relaxed structures, phonons, and reference electron-phonon couplings. $G_0W_0$ calculations with the Hybertsen-Louie plasmon-pole approximation supplied quasiparticle energies. GWPT added the difference between the GW self-energy variation and the DFPT exchange-correlation variation.

Electron-phonon Wannier interpolation placed energies, velocities, and couplings on dense Brillouin-zone grids. The Jacobi iteration solved the linearized ab initio Boltzmann transport equation. Calculations used a carrier concentration of 1013 cm-3 to approximate the near-intrinsic limit. A 10-10 T field captured the linear response.

For each phonon wavevector, GWPT required 3N calculations of the GW self-energy variation. Here, N is the number of atoms in the unit cell. GaP and Si used 100³ grids, while GaAs used 300³. Diamond and 3C-SiC used 180³ and 200³ grids, respectively. Reported values were extrapolated to the infinite sampling limit.

The authors also checked the quasiparticle approximation. For diamond, the generalized plasmon-pole GW energies had a 0.014 eV mean absolute error relative to full-frequency GW. DFT energies differed from full-frequency GW by 0.28 eV.

Figure 2(a) compares drift and Hall mobilities at room temperature for nominally undoped or weakly doped single-crystal samples. Table S2 provides the underlying values. Selected Hall-mobility comparisons are shown below.

MaterialGWBTE Hall mobilityExperimental Hall mobility
GaP315256
3C-SiC1377890
Si15291430
Diamond11721030
GaAs104248500

All entries use cm2/Vs at 300 K. Each row pairs one literature measurement with its calculated value. The table does not summarize every measurement used in the paper.

The full GWBTE method produced an 11% mean absolute relative error. DFT+DFPT produced 106%, while GW+DFPT produced 42%. Leave-one-material-out tests kept the GWBTE error between 10.4% and 13.4%. Every result stayed below 15% and at least two times smaller than the errors from either comparator. The authors describe the gain over the hybrid scheme as fourfold.

The broader test used 66 experimental measurements spanning growth methods, temperatures, and drift or Hall measurements. Only measurements above 250 K were included because phonon scattering dominates over impurity scattering in that range. The mean absolute relative error rose to 24% when each theoretical value used its matching experimental temperature.

The authors state that experimental scatter contributes to this metric. Many diamond measurements probe photo-induced free carriers under non-equilibrium conditions and report higher mobility values. The 300 K diamond reference instead measures thermal carriers in weakly doped material. The 24% value therefore reflects theoretical and experimental uncertainty.

Across the benchmark, GW self-energy effects changed electron-phonon coupling matrix elements by as much as 80%. For GaAs, the imaginary Fan-Migdal self-energy near the conduction-band minimum was 3.3 times larger with GW bands and GWPT couplings. The GW band effective mass increased by 89%, while the electron-phonon matrix elements increased by 14%.

GaP showed a 43% average increase in matrix elements for states within 1.0 eV of the conduction-band bottom. The transport-relevant coupling rose by 28%, while effective mass fell by 12%. Corrections varied widely around the DFPT values, so a single rescaling factor could not reproduce their wavevector and energy dependence.

Band-structure improvements drive the main GaAs correction. Larger coupling corrections dominate for 3C-SiC, GaP, and diamond. Silicon changes little beyond DFT, which explains why earlier DFT-level silicon calculations already agreed closely with experiments.

GWPT is computationally demanding. The reported CPU runs on the Frontera supercomputer used 3,387,144 core hours in total. The diamond run used 1,596,336 core hours, compared with 57,524 for GaAs. The method scales as $N_k N_q N^5 \log N$. Here, N is the number of atoms, while $N_k$ and $N_q$ are grid sizes.

The calculations omit finite-temperature corrections to atomic structure, electron bands, and phonon frequencies. They also neglect thermal expansion. Long-range dipole and quadrupole corrections remained at the DFPT level. The authors expect a long-range GWPT Fröhlich interaction to improve interpolation for polar materials.

The self-energy relaxation-time approximation served only to separate band and coupling effects. All mobilities in the main accuracy benchmarks came from full iterative Boltzmann solutions. Only electron mobility was calculated. Hole mobility requires spin-orbit coupling, which was not implemented in the GWPT framework.

The method applies to weakly to moderately correlated materials. Strong correlations require higher-order diagrams beyond first-order GW. The constant-screening approximation neglects variation of the screened Coulomb interaction. The 11% error should not be treated as the method’s intrinsic statistical accuracy. The authors say broader and more diverse datasets are needed to establish that accuracy.

Diamond provides a direct case for evaluating the model’s material implications. At 300 K, GWBTE predicted a drift mobility of 1290 cm2/Vs. Table S2 lists experimental drift values of 1800, 1802, and 1940 cm2/Vs.

For diamond, 3C-SiC, and GaP, the many-body coupling correction lowered predicted mobility. A DFT-only calculation can therefore overestimate intrinsic diamond mobility. For GaAs, the band-structure correction was the main driver.

Device studies must also separate phonon scattering from impurity scattering. The 300 K benchmark covers nominally undoped or weakly doped single-crystal samples. It does not by itself predict doped-device mobility below the phonon-dominated range. Using the method for new device materials requires GW corrections to both bands and electron-phonon coupling. Correcting the bands alone can be misleading.

The paper’s material requirement is nominally undoped or weakly doped single-crystal diamond for phonon-limited electron transport. 6CCVD’s diamond materials catalog includes single-crystal diamond grown by MPCVD, which matches that crystal family. For device-specific material and fabrication choices, the linked engineering support page provides custom fabrication support.

View Original Abstract

Accurate prediction of carrier mobility is critical for the discovery and design of next-generation electronic materials. Despite sustained progress, state-of-the-art ab initio methods remain limited by the approximate treatment of electron-phonon interactions at the density functional theory level. Here, we demonstrate that incorporating many-body GW corrections to both the electronic band structure and electron—phonon couplings when solving the ab initio Boltzmann transport equation yields a mean absolute relative error of just 11% for electron mobilities across benchmark semiconductors, including Si, GaAs, GaP, diamond, and SiC. The common practice of neglecting GW corrections to the electron—phonon interaction can lead to mobility errors exceeding 50%. The present findings highlight the importance of many-body GW self-energy effects in carrier transport simulations, and provides fundamental insights into how many-body electron—phonon interactions govern charge transport in crystalline solids.

  1. 2022 - Quantum Physics of Semiconductor Materials and Devices [Crossref]