input

BigDFT Input File Generation

Help the user create a BigDFT input file. The input can be produced as a YAML file (input.yaml) for direct use by the bigdft executable, or as a Python script using PyBigDFT.Inputfiles and PyBigDFT.InputActions. Ask each question one at a time. Skip questions whose answers are obvious from context.

Input File Format

BigDFT uses YAML input files with a one-to-one correspondence to Python dictionaries. The top-level keys are:

Section	Purpose
dft	Core DFT parameters: grid, XC, convergence, spin
posinp	Atomic positions, unit cell, boundary conditions
geopt	Geometry optimization
md	Molecular dynamics
kpt	K-point sampling (periodic systems)
mix	SCF mixing / diagonalization
output	What to write to disk
perf	Performance tuning, linear scaling control
lin_general	Linear scaling general parameters
lin_basis	Linear scaling support function parameters
lin_kernel	Linear scaling density kernel parameters
lin_basis_params	Per-element support function parameters
chess	CheSS solver parameters (FOE, NTPoly)
psolver	Poisson solver, implicit solvent
tddft	Linear-response TDDFT
sic	Self-interaction correction
mode	Calculator backend (DFT, Lennard-Jones, SIRIUS, etc.)
ig_occupation	Atomic orbital occupancies for input guess
psppar.<Element>	Per-element pseudopotential data

Questions

1 -- Calculation type

What type of calculation do you want to run?
  1. Single-point energy
  2. Geometry optimization
  3. Molecular dynamics
  4. Band structure
  5. TDDFT (linear response)
  6. Other / custom

2 -- System definition

How would you like to define your system?
  1. Provide atomic positions inline (I'll help you format them)
  2. Use an existing XYZ or ASCII file (posinp.xyz / posinp.ascii)
  3. Use an existing YAML file as a starting point
  4. Build from PyBigDFT (Systems/Atoms API)

If they provide positions inline or an XYZ file, ask about boundary conditions:

What boundary conditions?
  1. Free (isolated molecule) -- no cell needed
  2. Periodic (bulk crystal) -- need unit cell vectors
  3. Surface (periodic in x,y; free in z) -- need cell
  4. Wire (periodic in z; free in x,y) -- need cell

For periodic systems, ask for unit cell vectors and whether positions are in reduced (fractional) or Cartesian coordinates.

3 -- Exchange-correlation functional

Which XC functional?
  1. LDA  (ixc: 1)
  2. PBE  (ixc: PBE)  [recommended default]
  3. PBE0 (hybrid, ixc: PBE0)
  4. HF   (Hartree-Fock, ixc: HF)
  5. Other (provide name or ABINIT ixc code)

If they choose PBE on a free-boundary system, ask:

Add Grimme D3 dispersion correction? (recommended for molecules)

4 -- Grid parameters

Grid spacing and extension:
  - hgrids: wavelet grid spacing in bohr (smaller = more accurate, more expensive)
    Typical: 0.35-0.45 for production, 0.5+ for quick tests
  - rmult: [coarse, fine] radii multipliers
    Typical: [5.0, 8.0] for production, [3.5, 4.5] for quick tests

Use defaults (hgrids=0.4, rmult=[5.0, 8.0]) or customize?

5 -- Spin and charge

Only ask if relevant (transition metals, radicals, charged systems):

System charge and spin:
  - Net charge (qcharge): 0 for neutral
  - Spin polarization (nspin): 1=none, 2=collinear, 4=non-collinear
  - Magnetic moment (mpol): total magnetic moment in Bohr magnetons

6 -- SCF parameters

For most users, defaults are fine. Only ask if they want to customize:

SCF convergence:
  - gnrm_cv: wavefunction gradient norm (default: 1e-4, use 1e-5 for forces)
  - itermax: max iterations (default: 50)
  - Scaling approach: cubic (default) or linear?

If linear scaling, ask about the method:

Linear scaling kernel method:
  1. DIAG (diagonalization, default)
  2. FOE (Fermi Operator Expansion -- true linear scaling)
  3. NTPoly (polynomial expansion via NTPoly library)
  4. DIRMIN (direct minimization)

7 -- Calculation-specific options

For geometry optimization (type 2):

Optimization method:
  1. FIRE (default, robust)
  2. SQNM (Stabilized Quasi-Newton, good for large systems)
  3. LBFGS (Limited-memory BFGS)
  4. DIIS
  5. Other
Max steps? (default: 50)

For molecular dynamics (type 3):

MD parameters:
  - Number of steps (mdsteps)
  - Time step in atomic units (default: 20.67 a.u. ~ 0.5 fs)
  - Initial temperature in K (default: 300)
  - Thermostat: none (NVE) or nose_hoover_chain (NVT)?
  - Wavefunction extrapolation: 0 (none), 2 (BOMD, recommended)

For band structure (type 4):

K-point setup:
  - Method: mpgrid (Monkhorst-Pack) or auto (by resolution)
  - For mpgrid: grid size [nx, ny, nz]
  - For auto: kptrlen (K-space resolution in bohr)
  - Band structure path: provide high-symmetry points or use defaults

For TDDFT (type 5):

TDDFT approach:
  1. TDA (Tamm-Dancoff approximation)
  2. full (full Casida)
How many virtual states? (default: 8)

8 -- Output options

What output do you need?
  - Write orbitals to disk? (format: binary, text, etsf)
  - Write charge density?
  - Cube files around Fermi level?
  - Verbosity level? (0-3, default: 2)

9 -- Pseudopotentials

Pseudopotentials:
  1. Use built-in HGH-K (default, no setup needed)
  2. Use built-in HGH-K with NLCC (set_psp_nlcc)
  3. Use Krack pseudopotentials (set_psp_krack, PBE or LDA)
  4. Use custom psppar files from a directory

10 -- Advanced options

Only ask if the user seems experienced or explicitly requests:

Any advanced options?
  - Implicit solvent (water, ethanol, mesitylene)
  - External electric field
  - GPU acceleration (CUDA or OpenCL)
  - K-point sampling for periodic systems
  - Custom per-element basis parameters (linear scaling)
  - Electronic temperature / smearing
  - Constrained DFT

Output: YAML Format

After collecting answers, generate input.yaml. Use comments to explain non-obvious choices.

Example: Simple Single-Point (Free BC, PBE)

dft:
  hgrids: 0.4
  ixc: PBE
  rmult: [5.0, 8.0]
  gnrm_cv: 1.e-4
posinp:
  units: angstroem
  positions:
  - O: [0.000, 0.000, 0.119]
  - H: [0.000, 0.757, -0.476]
  - H: [0.000, -0.757, -0.476]
  properties:
    format: xyz

Example: Geometry Optimization

dft:
  hgrids: 0.4
  ixc: PBE
  rmult: [5.0, 8.0]
  gnrm_cv: 1.e-5      # tighter for force accuracy
geopt:
  method: FIRE
  ncount_cluster_x: 50
  betax: 4.0
  frac_fluct: 1.0
posinp:
  units: angstroem
  positions:
  - N: [0.0, 0.0, 0.0]
  - H: [0.0, 1.0, 0.0]
  - H: [-0.87, -0.5, 0.0]
  - H: [0.87, -0.5, 0.0]

Example: Molecular Dynamics (NVT)

dft:
  hgrids: 0.4
  ixc: PBE
  rmult: [6.0, 8.0]
  gnrm_cv: 5.e-2       # looser for MD
  itermax: 40
md:
  mdsteps: 1000
  timestep: 20.67       # ~0.5 fs
  temperature: 300.0
  thermostat: nose_hoover_chain
  print_frequency: 10
  wavefunction_extrapolation: 2
posinp: positions.xyz   # external file

Example: Periodic Solid (Si Bulk)

dft:
  rmult: [6.0, 8.0]
  ixc: PBE
kpt:
  method: mpgrid
  ngkpt: [4, 4, 4]
posinp:
  units: atomic
  cell: [10.261, 10.261, 10.261]
  positions:
  - Si: [0.0, 0.0, 0.0]
  - Si: [0.5, 0.5, 0.0]
  - Si: [0.5, 0.0, 0.5]
  - Si: [0.0, 0.5, 0.5]
  - Si: [0.25, 0.25, 0.25]
  - Si: [0.75, 0.75, 0.25]
  - Si: [0.75, 0.25, 0.75]
  - Si: [0.25, 0.75, 0.75]
  properties:
    reduced: Yes

Example: Linear Scaling (Large System)

dft:
  hgrids: 0.4
  rmult: [5.0, 7.0]
  gnrm_cv: 1.e-4
  itermax: 100
  inputpsiid: linear
  disablesym: Yes
lin_general:
  nit: [2, 3]
  rpnrm_cv: 1.e-6
lin_basis:
  idsx: [5, 0]
  gnrm_cv: [4.e-2, 1.e-2]
lin_kernel:
  linear_method: FOE
  rpnrm_cv: 1.e-8
lin_basis_params:
  C:
    ao_confinement: 2.7e-2
    confinement: 2.7e-2
    rloc: 4.7
    rloc_kernel: 8.0
    rloc_kernel_foe: 15.0
  H:
    nbasis: 1
    ao_confinement: 4.9e-2
    confinement: 4.9e-2
    rloc: 4.0
    rloc_kernel: 8.0
    rloc_kernel_foe: 15.0
chess:
  foe:
    ef_interpol_det: 1.e-12
    fscale: 5.e-2
posinp:
  units: angstroem
  positions:
  - C: [0.0, 0.0, 0.0]
  # ... more atoms

Example: Spin-Polarized (Transition Metal)

dft:
  ixc: PBE
  nspin: 2             # collinear spin
  mpol: 5              # magnetic moment in Bohr magnetons
  gnrm_cv: 5.e-4
  itermax: 20
  nrepmax: 2           # re-diagonalizations (or use 'accurate')
posinp:
  positions:
  - Mn: [0.0, 0.0, 0.0]
  - Mn: [0.0, 0.0, 5.0]

Example: Non-Collinear Spin

dft:
  nspin: 4             # non-collinear
  gnrm_cv: 5.e-4
  disablesym: Yes

Example: Dispersion Correction

dft:
  hgrids: 0.4
  ixc: PBE (ABINIT)    # ABINIT flavor needed for some dispersion
  dispersion: 5         # Grimme D3
  disablesym: Yes

Example: Implicit Solvent

dft:
  hgrids: 0.4
  ixc: PBE
psolver:
  environment:
    cavity: soft-sphere
    epsilon: 78.36       # water
    fact_rigid: 1.18
    delta: 0.625
    cavitation: Yes
    gammaS: 72.0
    alphaS: -60.5
    betaV: 0.0
    itermax: 20
    minres: 1.e-4

Example: With Custom Pseudopotentials

dft:
  ixc: PBE
psppar.O:
  Pseudopotential type: HGH-K + NLCC
  Atomic number: 8
  No. of Electrons: 6
  Pseudopotential XC: -101130
  Local Pseudo Potential (HGH convention):
    Rloc: 0.3455
    Coefficients (c1 .. c4): [-11.744, 1.907, 0.0, 0.0]
  Non Linear Core Correction term:
    Rcore: 0.345
    Core charge: 9.021
  NonLocal PSP Parameters:
  - Channel (l): 0
    Rloc: 0.368
    h_ij terms: [10.859, -0.430, 0.0, -2.129, 0.0, 0.0]

Example: Atomic Occupation Control

ig_occupation:
  Si:
    3s: 2.0
    3p: [0.667, 0.667, 0.667]
    3d: 0.0

Example: GPU Acceleration

perf:
  accel: OCLGPU         # OpenCL GPU acceleration
# or for CUDA:
psolver:
  setup:
    accel: CUDA

Output: Python Format

If the user prefers Python, generate a script using PyBigDFT.

Python Template

from BigDFT.Inputfiles import Inputfile
from BigDFT import InputActions as A

# Create input
inp = Inputfile()

# FILL: Grid parameters
inp.set_hgrid(FILL)          # e.g. 0.4
inp.set_rmult(coarse=FILL, fine=FILL)  # e.g. 5.0, 8.0

# FILL: XC functional
inp.set_xc('FILL')           # e.g. 'PBE'

# FILL: SCF convergence
inp.set_scf_convergence(gnrm=FILL)  # e.g. 1e-4

# FILL: Calculation-specific setup
# For geometry optimization:
# inp.optimize_geometry(method='FIRE', nsteps=50)

# FILL: Optional features
# inp.spin_polarize(mpol=1)
# inp.charge(charge=-1)
# inp.set_dispersion_correction()
# inp.set_implicit_solvent(solvent='water')
# inp.apply_electric_field([0, 0, 1e-3])
# inp.set_electronic_temperature(kT=1e-3)
# inp.use_gpu_acceleration(flavour='CUDA')
# inp.set_kpt_mesh('mpgrid', ngkpt=[4, 4, 4])

# FILL: Pseudopotentials (optional)
# inp.set_psp_krack('PBE')
# inp.set_psp_nlcc()

# FILL: Output
# inp.write_orbitals_on_disk(format='binary')
# inp.write_density_on_disk()

# FILL: Linear scaling (if needed)
# inp.set_linear_scaling()
# inp.set_ntpoly(thresh_dens=1e-6, conv_dens=1e-4)

# Atomic positions
inp.set_atomic_positions(FILL)  # posinp dict or use Systems API

# Print resulting YAML
import yaml
print(yaml.dump(dict(inp), default_flow_style=False))

Python Example: Full Workflow with SystemCalculator

Always do a dry run first to validate input and estimate memory before running the actual calculation. This catches errors like missing pseudopotentials, wrong atom counts, or insufficient memory before you wait in a queue.

from BigDFT.Inputfiles import Inputfile
from BigDFT.Calculators import SystemCalculator

# Create input
inp = Inputfile()
inp.set_xc('PBE')
inp.set_hgrid(0.4)
inp.set_rmult(coarse=5.0, fine=8.0)
inp.set_scf_convergence(gnrm=1e-4)

# Positions
posinp = {
    'units': 'angstroem',
    'positions': [
        {'O': [0.0, 0.0, 0.119]},
        {'H': [0.0, 0.757, -0.476]},
        {'H': [0.0, -0.757, -0.476]},
    ]
}
inp.set_atomic_positions(posinp)

# Dry run first -- validates input, estimates memory, does NOT run the calculation
calc_dry = SystemCalculator(dry_run=True)
log_dry = calc_dry.run(input=inp, name='water', run_dir='.')
# log_dry.energy is NaN (no calculation performed)
# Check the Memory Consumption Report in log_dry.log

# If dry run succeeds, run for real
calc = SystemCalculator()
log = calc.run(input=inp, name='water', run_dir='.')
print('Energy:', log.energy)
print('Forces:', log.forces)

The dry run uses bigdft-tool -a memory-estimation internally. It parses the input, sets up the grid, and reports memory requirements without performing the SCF. This is especially important before submitting to HPC queues via RemoteManager.

Python Example: Remove an Action

from BigDFT import InputActions as A

# Add then remove geometry optimization
inp.optimize_geometry(method='FIRE', nsteps=50)
inp.remove(A.optimize_geometry)

Variable Quick Reference

dft Section

Variable	Default	Description
hgrids	0.45	Grid spacing (bohr). Scalar or [hx, hy, hz].
rmult	[5.0, 8.0]	[coarse, fine] radius multipliers
ixc	1 (LDA)	XC functional. Use string names: PBE, LDA, HF, PBE0
gnrm_cv	1e-4	SCF convergence (gradient norm). Use 1e-5 for forces.
itermax	50	Max wavefunction optimization steps per cycle (subspace diag at end)
nrepmax	1	Number of cycles (each = itermax optimization + subspace diag). See SCF cycle structure below.
ncong	6	CG preconditioning iterations
idsx	6	DIIS history length
qcharge	0	System charge
nspin	1	1=unpolarized, 2=collinear, 4=non-collinear
mpol	0	Magnetic moment (Bohr magnetons)
inputpsiid	0	Input guess: 0=default, linear=linear scaling
dispersion	0	0=none, 5=Grimme D3
elecfield	[0,0,0]	Electric field (Ha/Bohr)
output_denspot	0	Output density: 0=none, 21=write after SCF
norbv	0	Virtual orbitals to compute
nvirt	0	Virtual orbitals to converge
nplot	0	Orbitals to plot as cube files
disablesym	No	Disable symmetry detection
alpha_hf	-1.0	Exact exchange fraction (hybrids)

geopt Section

Variable	Default	Description
method	none	FIRE, SQNM, LBFGS, BFGS, DIIS, SDCG, VSSD, PBFGS, NEB, SOCK, MD, LOOP
ncount_cluster_x	50	Max force evaluations
betax	4.0	Step size
frac_fluct	1.0	Force fluctuation fraction for convergence
forcemax	0	Max force criterion (Ha/Bohr)
randdis	0	Random displacement amplitude
nhistx	10	SQNM/SBFGS history length
trustr	0.5	Max single-atom displacement (SQNM)

md Section

Variable	Default	Description
mdsteps	0	Number of MD steps
timestep	20.67	Time step in a.u. (20.67 ~ 0.5 fs)
temperature	300	Initial temperature (K)
thermostat	none	none (NVE) or nose_hoover_chain (NVT)
print_frequency	1	Energy output frequency
wavefunction_extrapolation	0	0=none, 2=BOMD (recommended)
nose_frequency	3000	Thermostat frequency (cm^-1)
nose_chain_length	3	Nose-Hoover chain length
restart_pos / restart_vel / restart_nose	No	Restart flags

kpt Section

Variable	Default	Description
method	manual	auto, mpgrid, manual
ngkpt	[1,1,1]	Monkhorst-Pack grid (with method: mpgrid)
kptrlen	0	K-space resolution in bohr (with method: auto)
bands	No	Band structure calculation

mix Section

Variable	Default	Description
iscf	0	Mixing scheme. 0=direct minimization, 17=Pulay on density
norbsempty	0	Extra empty (virtual) bands to include. See SCF cycle structure below.
tel	0	Electronic temperature (Ha) for smearing
rpnrm_cv	1e-4	Residue convergence
alphamix	0	Mixing parameter

posinp Section

Field	Description
units	angstroem, atomic (bohr), or reduced (fractional)
cell	[a, b, c] for orthorhombic; omit for free BC
abc	Full cell vectors [[ax,ay,az],[bx,by,bz],[cx,cy,cz]]
positions	List of {Element: [x, y, z]} dicts
properties.reduced	Yes if using fractional coordinates
Can also be a string: filename of external XYZ/ASCII file

psolver.environment Section (Implicit Solvent)

Variable	Default	Description
cavity	none	none, soft-sphere, sccs
epsilon	78.36	Dielectric constant (78.36=water, 24.85=ethanol)
fact_rigid	1.12	Cavity size multiplier
delta	0.5	Transition region amplitude
gammaS	72.0	Surface tension (dyn/cm)
alphaS	-22.0	Repulsion free energy
betaV	-0.35	Dispersion free energy (GPa)

chess Section

Subsection	Key Variables
foe	fscale (decay length), eval_range_foe (eigenvalue bounds), accuracy_foe, ef_interpol_det
ntpoly	threshold_density, convergence_density, threshold_overlap, convergence_overlap, solver
lapack	blocksize_pdsyev, blocksize_pdgemm

lin_general Section

Variable	Default	Description
hybrid	No	Hybrid LS/CS mode
nit	[100, 100]	[low_accuracy, high_accuracy] iterations
rpnrm_cv	[1e-12, 1e-12]	Convergence thresholds
calc_dipole	No	Calculate dipole moment
kernel_restart_mode	0	Restart method (0=fresh, 1=from kernel, ...)
extra_states	0	Extra states for optimization
charge_multipoles	0	0=no, 1=Loewdin, 11=Mulliken

lin_basis Section

Variable	Default	Description
nit	[4, 5]	Support function optimization iterations
idsx	[6, 6]	DIIS history [low, high]
gnrm_cv	[1e-2, 1e-4]	Convergence [low, high]
fix_basis	1e-10	Fix SFs below this density change

lin_kernel Section

Variable	Default	Description
linear_method	DIAG	DIAG, FOE, NTPOLY, DIRMIN
nit	[5, 5]	Kernel iterations [low, high]
rpnrm_cv	[1e-10, 1e-10]	Convergence [low, high]
alphamix	[0.5, 0.5]	Mixing parameter [low, high]

lin_basis_params Section (Per-Element)

Variable	Default	Description
nbasis	4	Support functions per atom (1=s, 4=sp, 9=spd, 16=spdf)
ao_confinement	8.3e-3	Input guess confinement prefactor
confinement	[8.3e-3, 0.0]	Confinement [low, high accuracy]
rloc	[7.0, 7.0]	Localization radius [low, high]
rloc_kernel	9.0	Density kernel cutoff
rloc_kernel_foe	14.0	FOE matrix-vector cutoff

Typical per-element values (for organic molecules with FOE):

• C, N, O: nbasis=4, ao_confinement=2.7e-2, rloc=4.7, rloc_kernel=8.0, rloc_kernel_foe=15.0
• H: nbasis=1, ao_confinement=4.9e-2, rloc=4.0, rloc_kernel=8.0, rloc_kernel_foe=15.0
• Si: nbasis=9, ao_confinement=4.0e-2, rloc=6.0, rloc_kernel=7.0, rloc_kernel_foe=20.0

Profiles (Presets)

BigDFT supports importing profiles via the import key. Available built-in profiles:

Profile	Purpose
linear	Basic linear scaling (PDoS, charge analysis, MD)
linear_accurate	High-accuracy linear scaling with FOE
linear_moderate	Balanced accuracy/speed
linear_fast	Speed over accuracy
linear_fragments	Fragment calculations
linear_purify	NTPoly purification instead of CheSS
mixing	Metallic systems with density mixing

Usage in YAML:

import: linear

Usage in Python:

inp.load(profile='linear')

SCF Cycle Structure: itermax, nrepmax, and Virtual Orbitals

BigDFT's SCF works in cycles. Understanding itermax, nrepmax, and norbsempty together is essential for hard-to-converge systems and for computing virtual orbitals.

The basic loop

for irep = 1 to nrepmax:
    for iter = 1 to itermax:
        optimize wavefunctions (gradient descent / DIIS)
        check gnrm_cv convergence → break if converged
    end
    subspace diagonalization  ← reassigns occupations, mixes orbitals
end

Each cycle consists of up to itermax wavefunction optimization steps followed by a subspace diagonalization. The subspace diagonalization builds the Hamiltonian matrix in the current orbital basis, diagonalizes it, and reorders/mixes the orbitals by eigenvalue. This is where occupation numbers get properly assigned.

The process repeats for nrepmax cycles.

Default behavior (no virtual orbitals)

By default, norbsempty = 0 and nrepmax = 1. BigDFT only optimizes occupied orbitals, runs one cycle of up to itermax steps, does a final subspace diagonalization, and stops. This is sufficient for most ground-state calculations.

Why you need virtual orbitals

There are two reasons to set norbsempty > 0 in the mix section:

1. Hard-to-converge systems (near-degenerate HOMO-LUMO, metals, open-shell): Including extra virtual orbitals in the optimization lets the subspace diagonalization properly shuffle electrons between nearly degenerate levels. Without them, the optimizer can oscillate because it doesn't have room to move electrons into the right orbitals.

2. You want to inspect virtual orbitals: For post-processing (band gaps, spectral properties, etc.), you need unoccupied orbitals. Setting norbsempty produces them.

Why nrepmax matters with virtuals

With virtual orbitals, nrepmax > 1 is important because:

• After the first cycle's subspace diagonalization, occupation numbers may change (electrons move between orbitals that were near the Fermi level)
• The newly reordered orbitals need further optimization in the next cycle
• Each cycle refines both the orbitals and their occupations

For metallic or near-degenerate systems, use nrepmax: accurate (which auto-selects up to 10 cycles) or set it manually:

dft:
  gnrm_cv: 1.e-5
  itermax: 50
  nrepmax: 5
mix:
  norbsempty: 10
  tel: 0.001       # small smearing helps convergence

Example: Metallic system with smearing

import: mixing
dft:
  ixc: PBE
  hgrids: 0.4
  itermax: 50
  nrepmax: 10
mix:
  iscf: 17          # Pulay mixing on density
  norbsempty: 20    # extra bands for the Fermi surface
  tel: 0.01         # electronic temperature for smearing
  alphamix: 0.5

Example: Getting virtual orbitals for a molecule

dft:
  ixc: PBE
  hgrids: 0.35
  gnrm_cv: 1.e-5
  itermax: 100
  nrepmax: 3          # enough cycles to settle occupations
mix:
  norbsempty: 5       # 5 virtual orbitals

Notes

• For free boundary conditions (isolated molecules), omit cell and kpt.
• For periodic systems, always specify cell (or abc) and kpt.
• gnrm_cv: accurate is a valid shortcut that auto-selects a tight threshold.
• nrepmax: accurate auto-selects re-diagonalization count.
• posinp can be a string pointing to an external file: posinp: myfile.xyz
• Positions are formatted as - Element: [x, y, z] in YAML.
• Linear scaling (inputpsiid: linear) requires disablesym: Yes and per-element lin_basis_params.
• The (ABINIT) suffix on ixc (e.g., PBE (ABINIT)) selects the ABINIT flavor of the functional, needed for some features like dispersion.
• Units: BigDFT works internally in atomic units (bohr, hartree). Positions can use angstroem or atomic.
• Time unit: 1 a.u. of time ~ 0.02419 fs. A timestep of 20.67 a.u. ~ 0.5 fs.
• Always validate with a dry run (SystemCalculator(dry_run=True)) before submitting real calculations, especially on HPC systems. It catches input errors and estimates memory in seconds.