Home › Resources › Tools Directory › 1. GROUND-STATE DFT › 1.9 Real-Space

RSDFT (Real-Space Density Functional Theory)

RSDFT is a high-performance ab initio density functional theory code based on the real-space finite-difference method. It is explicitly designed for massively parallel computing architectures, enabling first-principles calculations on sy…

1. GROUND-STATE DFT 1.9 Real-Space VERIFIED
Back to Directory Official Website

Overview

RSDFT is a high-performance ab initio density functional theory code based on the real-space finite-difference method. It is explicitly designed for massively parallel computing architectures, enabling first-principles calculations on systems containing over 100,000 atoms. The code avoids the global communication bottlenecks of Fast Fourier Transforms (FFTs) typical in plane-wave codes, making it highly scalable on supercomputers like the K computer and Fugaku.

Reference Papers

Reference papers are not yet linked for this code.

Full Documentation

Official Resources

  • Homepage: http://rsdft.jp/
  • Source Repository: https://github.com/j-iwata/RSDFT
  • License: Open Source (GPLv3)

Overview

RSDFT is a high-performance ab initio density functional theory code based on the real-space finite-difference method. It is explicitly designed for massively parallel computing architectures, enabling first-principles calculations on systems containing over 100,000 atoms. The code avoids the global communication bottlenecks of Fast Fourier Transforms (FFTs) typical in plane-wave codes, making it highly scalable on supercomputers like the K computer and Fugaku.

Scientific domain: Large-scale materials science, Nanostructures, Surfaces, Interfaces Target user community: HPC researchers, large-scale DFT practitioners

Theoretical Methods

  • Real-Space Finite-Difference Method
  • Kohn-Sham Density Functional Theory
  • Pseudopotentials (Norm-conserving and Ultrasoft)
  • Real-space grid discretization
  • High-order finite difference stencils (e.g., higher-order central differences)
  • Conjugate Gradient and RMM-DIIS eigensolvers
  • Projector Augmented Wave (PAW) method (experimental/supported in versions)

Capabilities

  • System Size: Routine calculations for 10,000 to 100,000+ atoms.
  • Parallelism: Scalable to >80,000 nodes (Gordon Bell Prize Winner 2011).
  • Electronic Structure: Ground state energy, forces, stresses.
  • Molecular Dynamics: Born-Oppenheimer MD for large systems.
  • Boundary Conditions: Flexible (Isolated, Periodic, Wire, Slab).

Key Strengths

Massively Parallel Scalability:

  • No FFT bottleneck.
  • Domain decomposition allows linear scaling with respect to processors for fixed system size per core.
  • Demonstrated on exascale-class hardware.

Large-Scale Calculations:

  • Can handle silicon nanowires with 100k+ atoms.
  • Ideal for complex nanostructures where periodic boundary conditions of plane-waves are artificial.

Flexible Boundary Conditions:

  • Naturally handles non-periodic systems without supercell approximation artifacts (charged systems, dipoles).

Inputs & Outputs

  • Inputs:
    • Grid spacing
    • Finite difference order
    • Atomic coordinates
    • Pseudopotential files
  • Outputs:
    • Total energies and forces
    • Charge density distribution (cube files)
    • Wavefunctions (real-space grid)

Interfaces & Ecosystem

  • Python: Prototyping environment available in Python.
  • HPC Integration: Optimized for MPI/OpenMP hybrid parallelism.
  • MateriApps: Listed and supported via the MateriApps ecosystem.

Advanced Features

  • GPU Acceleration: Ports available for GPU-accelerated clusters.
  • Order-N capability: Methods for linear scaling electronic structure (Chebyshev filtering).

Performance Characteristics

  • Speed: Superior to Plane-Wave codes for very large systems (>1000 atoms) on massive core counts.
  • Accuracy: Systematic convergence with grid spacing (h) and stencil order.
  • Efficiency: Excellent weak scaling.

Computational Cost

  • High: For small systems (under 100 atoms), overhead is higher than VASP/QE.
  • Low: For large systems (>5000 atoms), significantly cheaper/faster than PW codes.

Limitations & Known Constraints

  • Pseudopotentials: requires real-space optimized potentials for best efficiency.
  • Grid anisotropy: "Egg-box" effect possible if grid is too coarse (breaking translation symmetry).
  • Documentation: English documentation can be sparse compared to VASP; website is partly in Japanese.

Comparison with Other Codes

  • vs PARSEC: Both are real-space; RSDFT focuses more on massive HPC scalability.
  • vs ONETEP: ONETEP attempts O(N) via Wannier functions; RSDFT uses domain decomposition of the grid.
  • vs Plane-Wave: RSDFT eliminates FFT communication, winning at extreme scale.
  • Unique strength: Gordon Bell Prize-winning scalability for 100k atom systems.

Application Areas

  • Nanowires: Electronic properties of realistic diameter wires.
  • Biomolecules: Large proteins in solvent (implicit/explicit).
  • Defects: Dilute defects in very large supercells.

Best Practices

  • Grid Convergence: Test h-grid spacing carefully to avoid egg-box errors.
  • Parallel Layout: Match domain decomposition to physical system shape.
  • Pseudopotentials: Use soft potentials where possible to allow coarser grids.

Community and Support

  • MateriApps: Integration with the Japanese materials science setup.
  • Development Team: Based at U-Tokyo and RIKEN.

Verification & Sources

Primary sources:

  1. Official Website: http://rsdft.jp/
  2. GitHub: https://github.com/j-iwata/RSDFT
  3. J. Hasegawa et al., "First-Principles Calculations of 100,000-Atom Silicon Nanowires", SC11 (Gordon Bell Prize).

Verification status: ✅ VERIFIED

  • Source code: OPEN (GPLv3)
  • Performance: Confirmed high-impact HPC code.

Related Tools in 1.9 Real-Space

M-SPARC RSDFT (Real-Space Density Functional Theory)