Official Resources
- Homepage: https://abacus.deepmodeling.com/
- Documentation: https://abacus.deepmodeling.com/en/latest/
- Source Repository: https://github.com/deepmodeling/abacus-develop
- License: GNU Lesser General Public License v3.0
Overview
ABACUS (Atomic-orbital Based Ab-initio Computation at UStc) is an open-source DFT package supporting both plane-wave and numerical atomic orbital (NAO) basis sets. Developed by the DeepModeling community in China, ABACUS is designed for efficient electronic structure calculations, materials simulations, and integration with machine learning workflows. It features excellent GPU acceleration, hybrid basis sets, and is particularly strong in Chinese academic and industrial applications.
Scientific domain: DFT with multiple basis sets, materials science, machine learning integration
Target user community: Materials scientists, ML researchers, Chinese research community
Theoretical Methods
- Kohn-Sham DFT (LDA, GGA, meta-GGA)
- Plane-wave basis sets
- Numerical atomic orbitals (NAO)
- Hybrid plane-wave/NAO basis
- Norm-conserving pseudopotentials
- Stochastic DFT methods
- van der Waals corrections (DFT-D2/D3)
- DFT+U for correlated systems
- Hybrid functionals
- Spin-orbit coupling
- Non-collinear magnetism
Capabilities (CRITICAL)
- Ground state electronic structure
- Geometry optimization
- Molecular dynamics (Born-Oppenheimer)
- Cell relaxation
- Band structures and DOS
- Phonons via DFPT
- Elastic constants
- Surface calculations
- Machine learning interface
- DeepKS (ML correction)
- GPU acceleration
- Multiple basis sets (PW, NAO, hybrid)
- Stochastic DFT for large systems
- Linear-scaling algorithms
- Excellent parallelization
- Python interface (PyABACUS)
Sources: Official ABACUS documentation (https://abacus.deepmodeling.com/), GitHub repository
Key Strengths
Multiple Basis Sets:
- Plane-waves (PW)
- Numerical atomic orbitals (NAO)
- Hybrid PW/NAO
- Flexibility in choice
- Method comparison
Machine Learning:
- DeepModeling integration
- DeepKS correction
- ML potential interface
- DP-GEN workflows
- Modern AI integration
GPU Acceleration:
- CUDA support
- Significant speedup
- Production-ready
- Optimized kernels
Stochastic DFT:
- Large system capability
- Linear scaling
- Statistical sampling
- Thousands of atoms
Open Source:
- LGPL v3 licensed
- Active GitHub development
- Growing community
- Chinese and international
Inputs & Outputs
-
Input formats:
- INPUT file (main parameters)
- STRU file (structure)
- KPT file (k-points)
- Pseudopotential files
-
Output data types:
- Text output (OUT.*)
- Structure files
- Wavefunction data
- Band structure files
- DOS files
Interfaces & Ecosystem
-
DeepModeling:
- DP-GEN workflows
- DeePMD integration
- DeepKS correction
- ML ecosystem
-
Analysis:
- PyABACUS (Python interface)
- Custom tools
- Standard visualization
-
Parallelization:
- MPI parallelization
- GPU acceleration (CUDA)
- Hybrid parallelization
- Good scaling
Workflow and Usage
Example INPUT File:
INPUT_PARAMETERS
calculation scf
basis_type lcao
ecutwfc 100
scf_thr 1.0e-7
scf_nmax 100
smearing_method gaussian
smearing_sigma 0.001
kspacing 0.1
Example STRU File:
ATOMIC_SPECIES
Si 28.0855 Si_ONCV_PBE-1.0.upf
LATTICE_CONSTANT
5.43
LATTICE_VECTORS
0.5 0.5 0.0
0.5 0.0 0.5
0.0 0.5 0.5
ATOMIC_POSITIONS
Direct
Si
0.0
2
0.00 0.00 0.00 1 1 1
0.25 0.25 0.25 1 1 1
Running ABACUS:
abacus
mpirun -np 16 abacus
Advanced Features
Stochastic DFT:
- Statistical sampling
- Large systems (1000+ atoms)
- Linear-scaling approach
- Reduced computational cost
- Controlled accuracy
DeepKS:
- Machine learning correction
- Improved accuracy
- Trained models
- DeepModeling framework
- Enhanced DFT
Multiple Basis Options:
- Choose PW for accuracy
- Choose NAO for efficiency
- Hybrid for balance
- System-dependent optimization
GPU Acceleration:
- Offload to GPU
- CUDA kernels
- Speedup factors
- Multi-GPU support
DFPT:
- Phonon calculations
- Dielectric response
- Born effective charges
- Efficient implementation
Performance Characteristics
- Speed: Competitive
- Scaling: Good parallel performance
- GPU: Significant acceleration
- Memory: Moderate
- Typical systems: 50-1000 atoms
Computational Cost
- DFT (PW): Standard
- DFT (NAO): More efficient
- Stochastic: Very efficient for large
- GPU: Dramatically faster
- ML corrections: Minimal overhead
Limitations & Known Constraints
- Smaller community: Less established globally
- Documentation: Good but primarily Chinese/English
- Learning curve: Moderate
- Features: Fewer than VASP/QE
- Platform: Linux primarily
- GPU: CUDA only
Comparison with Other Codes
- vs VASP: ABACUS open-source, multiple basis
- vs Quantum ESPRESSO: Similar PW capabilities, ABACUS adds NAO
- vs SIESTA: Both NAO, ABACUS also PW
- vs CP2K: Different implementations
- Unique strength: Multiple basis sets, ML integration, stochastic DFT, DeepModeling ecosystem
Application Areas
Materials Science:
- Electronic structure
- Structural properties
- Phase stability
- Defects and dopants
Machine Learning:
- Training data generation
- DeepKS applications
- ML potential development
- DP-GEN workflows
Large Systems:
- Stochastic DFT
- Amorphous materials
- Complex systems
- Interfaces
Method Development:
- Basis set comparison
- Algorithm testing
- ML integration
Best Practices
Basis Set Selection:
- PW for high accuracy
- NAO for large systems
- Test both for system
- Convergence testing
Convergence:
- Plane-wave cutoff (PW)
- NAO basis size
- K-point sampling
- SCF threshold
GPU Usage:
- Enable GPU acceleration
- Monitor utilization
- Balance CPU-GPU
- Optimize workload
ML Integration:
- Use DeepKS when appropriate
- Train models carefully
- Validate corrections
- Follow DeepModeling best practices
Community and Support
- Open-source (LGPL v3)
- Active GitHub repository
- Documentation website
- DeepModeling community
- Chinese and international users
- Regular updates
Educational Resources
- Online documentation
- Tutorial examples
- GitHub wiki
- DeepModeling resources
- Published papers
Development
- Active GitHub development
- DeepModeling project
- University of Science and Technology of China
- Community contributions
- Regular releases
DeepModeling Ecosystem
- Part of DeepModeling community
- Integration with DeePMD
- DP-GEN workflows
- ML potential development
- AI for science
Verification & Sources
Primary sources:
- Official website: https://abacus.deepmodeling.com/
- Documentation: https://abacus.deepmodeling.com/en/latest/
- GitHub repository: https://github.com/deepmodeling/abacus-develop
- M. Chen et al., Comput. Phys. Commun. 285, 108634 (2023) - ABACUS overview
Secondary sources:
- ABACUS documentation and tutorials
- DeepModeling community resources
- Published studies using ABACUS
- GitHub discussions
Confidence: VERIFIED - GitHub repository and official website confirmed
Verification status: ✅ VERIFIED
- Official homepage: ACCESSIBLE
- Documentation: COMPREHENSIVE
- Source code: OPEN (GitHub, LGPL v3)
- Community support: GitHub issues, documentation
- Active development: Very active GitHub
- Specialized strength: Multiple basis sets (PW/NAO), ML integration, stochastic DFT, DeepModeling ecosystem, GPU acceleration