Poraquê

Poraquê is a compact, readable density-functional theory (DFT) code for electronic-structure calculations. It implements Kohn-Sham (KS-DFT), orbital-free (OF-DFT), and Frozen-Density Embedding (FDE / subsystem DFT) behind a single calculator, and integrates natively with the Atomic Simulation Environment (ASE) so that structures, workflows, and analysis tools from the wider ecosystem work out of the box.

Unlike a traditional monolithic DFT package, Poraquê is built around one shared real-space / plane-wave numerical core that is reused by every method. That design is what makes its three defining capabilities possible: a native ASE calculator as the only user-facing API, the ability to seamlessly bridge orbital-free and Kohn-Sham regions in a single system via freeze-and-thaw embedding, and a research focus on machine-learning-derived kinetic energy density functionals (ML-KEDFs).

What makes Poraquê different

Traditional DFT codes (Quantum ESPRESSO, VASP, ABINIT, GPAW, …) are powerful but typically commit to a single electronic-structure paradigm (almost always Kohn-Sham), expose a bespoke input-file language, and treat the kinetic energy functional as fixed. Poraquê is deliberately different:

• Native ASE integration, not a wrapper. The only public entry point is a standard ASE Calculator (poraque.ase.Poraque). There is no custom input language: you build an Atoms object and ask for get_potential_energy() / get_forces(). Every ASE tool — builders, optimizers, equation-of-state, databases, NEB — works unchanged. The method is chosen with a single keyword, mode='ks' or mode='of'.
• A hybrid OF ⇄ KS engine via Freeze-Embedding. Because OF-DFT and KS-DFT share the same grid, functionals, and Hartree/XC machinery, Poraquê can partition a system and treat different regions with different methods at the same time — an accurate Kohn-Sham active subsystem embedded in a cheap orbital-free environment (KS-in-OF), or any combination — coupled through a non-additive kinetic + electrostatic + XC embedding potential and relaxed with freeze-and-thaw cycles. This OF/KS bridging is awkward or impossible in codes built around a single paradigm.
• Machine-learning-derived KEDFs as a first-class research target. The orbital-free kinetic energy is treated as a pluggable, learnable object, not a hard-coded term. The functional interface (energy + functional derivative) is the same one used by Thomas-Fermi and von Weizsäcker, so a symbolic- regression formula or a CNN trained on electron-density slices can drop straight into the self-consistent minimizer.
• One readable core, in plain NumPy/SciPy. The reference implementation favors clarity over micro-optimization, with a clean Grid / System / Density / Result data model and a pluggable backend layer — a code you can actually read, extend, and use to prototype new physics.

	Traditional KS-DFT codes	Poraquê
Primary interface	Custom input files	Native ASE Calculator
Methods	Kohn-Sham only (usually)	KS-DFT + OF-DFT + FDE, one API
OF/KS in one system	Not supported	KS-in-OF / OF-in-KS embedding
Kinetic functional	Fixed	Pluggable, ML-derived KEDFs
Pseudopotentials	.upf (norm-conserving/PAW)	.upf (PseudoDojo/QE) + analytic
Codebase	Large Fortran/C++	Compact, readable Python

Features

• Unified calculator. One ASE calculator, poraque.ase.Poraque, selects the method dynamically with mode='ks' (Kohn-Sham) or mode='of' (orbital-free).
• Plane-wave basis (mandatory for KS-DFT) with automatic grid generation. Fields and orbitals live on a uniform real-space grid whose discrete Fourier transform spans a plane-wave basis. The kinetic operator is applied exactly in reciprocal space (½ |G + k|²); local potentials are applied diagonally in real space. Set a plane-wave cutoff with ecut and the real-space grid is sized automatically to resolve it (Nyquist condition); a manually supplied grid_shape is safely overridden by the optimized grid (with a warning).
• Pseudopotentials, including .upf. A modular poraque.pseudopotentials package provides a transparent core–valence split, built-in analytic local pseudopotentials, and a reader for .upf (Unified Pseudopotential Format) files compatible with PseudoDojo and Quantum ESPRESSO. A bundled registry maps the chosen exchange-correlation functional to the matching file (pseudopotentials='upf' picks LDA or PBE pseudopotentials to match xc).
• Transparent energy accounting. Results carry a strictly additive energy decomposition — total, kinetic, external/ionic, Hartree, exchange-correlation, and nonlocal terms — and an optional verbose=True mode prints the generated grid, the material structure (cell, positions, PBC), the per-step SCF convergence, and the final breakdown to standard output.
• Periodic systems & k-points. Full periodic boundary conditions with Brillouin-zone sampling via Monkhorst–Pack grids built on ase.dft.kpoints (kpts=(n1, n2, n3)), folded by time-reversal symmetry.
• Energies and forces reported in ASE units (eV, eV/Å), plus frozen-density embedding (FDE) drivers for subsystem calculations.

Installation

Poraquê targets Python ≥ 3.10. For development, clone the repository and install in editable mode:

git clone https://github.com/seixas-research/poraque.git
cd poraque
pip install -e .

This pulls in the runtime dependencies (NumPy, SciPy, ASE, pandas, matplotlib). Run the test suite with:

pytest

Quick start

The calculator is a drop-in ASE Calculator: attach it to an Atoms object and ask for energies or forces. The full calculation log — generated grid, material structure, step-by-step SCF convergence, and the final energy decomposition — is printed to standard output by default (pass verbose=False to silence it).

You control the plane-wave grid in one of two equivalent ways: pass an explicit grid_shape=(Nx, Ny, Nz), or pass a plane-wave kinetic-energy cutoff ecut (Hartree) and let Poraquê size the grid automatically. Grid.from_ecut exposes the exact mapping so you can see the dimensions a cutoff produces.

1. Gas-phase molecule — an H₂ molecule (orbital-free DFT)

from ase import Atoms

from poraque.ase import Poraque
from poraque.core import Grid, SolverSettings, System

# H2 molecule in a non-periodic box (Ångström).
h2 = Atoms(
    "H2",
    positions=[[2.0, 2.5, 2.5], [3.0, 2.5, 2.5]],
    cell=[5.0, 5.0, 5.0],
    pbc=False,
)

# A user-supplied kinetic-energy cutoff dynamically determines the grid.
ecut = 8.0  # Hartree
system = System.from_ase(h2)
derived = Grid.from_ecut(system.cell, ecut, pbc=system.pbc)
print(f"Ecut = {ecut} Ha  ->  grid_shape = {derived.shape}")  # e.g. (26, 26, 26)

h2.calc = Poraque(
    mode="of",                       # orbital-free DFT
    ecut=ecut,                       # grid generated automatically from the cutoff
    external_kwargs={"a": 0.8},      # softening of the nuclear potential
    settings=SolverSettings(max_iter=80, mixing=0.1),
)
# Equivalent explicit form: Poraque(mode="of", grid_shape=(26, 26, 26), ...)

print(f"Total energy: {h2.get_potential_energy():.6f} eV")  # also prints the log
print("Forces (eV/Å):")
print(h2.get_forces())

2. Bulk crystal — silicon with k-point sampling (Kohn-Sham DFT)

from ase.build import bulk

from poraque.ase import Poraque
from poraque.core import Grid, SolverSettings, System

# Diamond-structure silicon (2-atom primitive cell).
si = bulk("Si", "diamond", a=5.43)

# Size the plane-wave grid from a cutoff (KS-DFT always uses the plane-wave basis).
ecut = 6.0  # Hartree
derived = Grid.from_ecut(System.from_ase(si).cell, ecut)
print(f"Ecut = {ecut} Ha  ->  grid_shape = {derived.shape}")

si.calc = Poraque(
    mode="ks",                       # Kohn-Sham DFT (plane-wave basis, mandatory)
    ecut=ecut,                       # grid_shape derived automatically; see above
    kpts=(4, 4, 4),                  # Monkhorst-Pack Brillouin-zone sampling
    pseudopotentials="auto",         # 4 valence electrons per Si atom
    settings=SolverSettings(max_iter=40, mixing=0.5, tolerance=1e-5),
)
# Or specify the grid directly: Poraque(mode="ks", grid_shape=(16, 16, 16), ...)

print(f"Total energy: {si.get_potential_energy():.6f} eV")
print(f"Valence electrons: {si.calc.results['density'].integrate():.4f}")

If you pass both ecut and grid_shape, the cutoff wins: the manual grid is overridden by the automatically optimized one and a warning is logged.

More runnable scripts live in examples/.

Method selection at a glance

Argument	Meaning
mode	'ks' (Kohn-Sham) or 'of' (orbital-free)
basis	Single-particle basis; 'pw' (plane waves), mandatory for KS-DFT
grid_shape	Real-space grid (Nx, Ny, Nz) (overridden when ecut is set)
ecut	Plane-wave cutoff (Hartree); sizes the grid automatically
kpts	(n1, n2, n3) Monkhorst-Pack grid, or explicit fractional k-points
pseudopotentials	'auto' (analytic), 'upf' (bundled UPF), a {symbol: spec} mapping, or a LocalPseudopotential
pseudo_functional	Functional for UPF selection ('LDA'/'PBE'); inferred from xc if unset
xc	Exchange-correlation functional ('lda' by default, None to disable)
charge	Net charge of the system
verbose	Print grid, structure, SCF convergence, and energy decomposition

License

This is an open source code under the MIT License.