ABINIT-GW

Official Resources

• Homepage: https://www.abinit.org/
• Documentation: https://docs.abinit.org/theory/mbt/
• Source Repository: https://github.com/abinit/abinit
• License: GNU General Public License

Overview

ABINIT-GW refers to the many-body perturbation theory (MBPT) functionality within the ABINIT package, an open-source plane-wave pseudopotential code for electronic structure calculations. ABINIT implements the GW approximation for computing quasiparticle energies and band gaps, providing one of the major open-source GW implementations alongside BerkeleyGW and Yambo.

Within MBPT, ABINIT calculates quasiparticle (QP) energies and amplitudes by solving a nonlinear equation involving the non-Hermitian, nonlocal, and frequency-dependent self-energy operator. A typical GW calculation in ABINIT consists of two steps following a DFT calculation: first computing the screened interaction (dielectric matrix) and storing it on disk (optdriver=3), then evaluating the self-energy matrix elements to obtain QP corrections (optdriver=4). ABINIT supports plasmon-pole models and full numerical frequency integration, with the latter using contour integration along the imaginary axis.

Scientific domain: Many-body perturbation theory, quasiparticle band structure
Target user community: Computational materials scientists requiring open-source GW calculations

Theoretical Methods

• GW approximation within Hedin's equations
• G0W0 (single-shot quasiparticle energies)
• Self-consistent GW (scGW) and GW0 (self-consistent G, fixed W0)
• RPA screened interaction computation
• Plasmon-pole models (Godby-Needs, etc.)
• Full numerical frequency integration (contour deformation)
• GW 1-body reduced density matrix (1RDM) with Galitskii-Migdal correlation
• PAW and norm-conserving pseudopotential frameworks

Capabilities (CRITICAL)

• Quasiparticle energy calculations (G0W0, scGW, GW0)
• Screening (susceptibility and dielectric matrix) computation
• Self-energy matrix element evaluation
• Multiple frequency integration methods (plasmon-pole and contour)
• Coulomb singularity treatment (icutcoul) for convergence acceleration
• Support for PAW and norm-conserving pseudopotentials
• Core contribution to self-energy via Fock operator (PAW)
• GW total energy via Galitskii-Migdal approximation
• BSE calculations (optdriver=99) on top of GW

Inputs & Outputs

Input formats:

• ABINIT input files with optdriver=3 (screening) and optdriver=4 (self-energy)
• WFK files from preceding DFT calculation
• SCR files (screening/dielectric matrix)

Output data types:

• Quasiparticle energies and corrections
• Self-energy matrix elements (exchange and correlation parts)
• Screened interaction (SCR files)
• Dielectric matrices
• GW total energies (with Galitskii-Migdal)

Interfaces & Ecosystem

• Programming language: Fortran (with Python bindings)
• Open-source: GNU GPL license
• Parallel computing: MPI and OpenMP parallelization
• Prerequisite: ABINIT DFT ground-state calculation
• Downstream: BSE calculations using SCR files
• Parameter files: .ac9 files for compilation options (enable_gw_dpc)

Limitations & Known Constraints

• Pseudopotential-based (core-valence interaction approximated)
• Memory-intensive for large dielectric matrices
• Convergence with k-points, unoccupied states, and frequency grid required
• Self-consistent GW variants are computationally very expensive

Performance Characteristics

• Screening computation: scales with number of bands and G-vectors (ecuteps)
• Self-energy: controlled by ecutsigx (exchange) and ecuteps (correlation)
• FFT-based oscillator matrix evaluation (fftgw parameter)
• Memory reduction option: enable_gw_dpc="no" at compilation
• Parallel over k-points and frequency points

Comparison with Other Codes

• vs VASP-GW: ABINIT is open-source (GPL); VASP is commercial. Both offer comprehensive GW
• vs BerkeleyGW: BerkeleyGW is standalone post-processing; ABINIT-GW is integrated
• vs Yambo: Yambo works with QE/ABINIT output; ABINIT-GW is self-contained
• vs FHI-gap: FHI-gap is all-electron LAPW; ABINIT uses pseudopotentials

Best Practices

• Converge ecuteps (screening) and ecutsigx (self-energy) separately
• Use sufficient empty bands (nband) for correlation part of self-energy
• For metals, use contour deformation instead of plasmon-pole
• Use icutcoul for improved k-point convergence of Coulomb singularity
• Follow GW1 and GW2 tutorials for systematic convergence testing

Verification & Sources

Primary sources:

1. ABINIT MBPT theory: https://docs.abinit.org/theory/mbt/
2. ABINIT GW topic: https://docs-10-4-3.abinit.org/topics/GW/
3. ABINIT GW tutorials: https://docs.abinit.org/tutorial/gw1/

Confidence: VERIFIED - Official documentation and tutorials confirmed accessible