This demo initializes a Cartesian potential magnetic field from a Python V1 data-driven boundary frame.

Expected boundary-frame layout:

snapshot_time, nx, ny, dx, dy, Bx, By, Bz

dx and dy are stored in km, and magnetic-field components are stored in Gauss. The potential-field extrapolation uses the Bz component.

Two extrapolation methods are available through usr_list:

potential_field_method='fft'   ! default
fft_padding_factor=2
fft_top_boundary='open'         ! 'open' or 'closed'
lalpha=0.d0                     ! nonzero for a constant-alpha FFT LFFF
lfff_flux_treatment='strict'    ! or 'subtract_mean'
lfff_max_flux_imbalance=0.1d0

The default ‘potential_field_method='fft’uses the dependency-free spectral solver. Setpotential_field_method='green'to use the legacy Green-function solver. The data-driven geometry places the magnetogram atz=0on the first lower ghost-cell center. The physical lower face is therefore half a finest-cell spacing above zero. The FFT solver uses the same half-cell source-plane depth, and requiresllift=0`. A padding factor of 1 is a strictly periodic horizontal solution; the default factor of 2 centers the magnetogram in a zero-padded plane to reduce periodic-image effects.

For ‘fft_top_boundary='open’, every nonzero mode decays into a half-space. A constant-alpha LFFF in this mode requires a flux-balanced magnetogram and abs(lalpha)smaller than the lowest nonzero wavenumber of the padded plane. Forfft_top_boundary='closed', the nonzero modes satisfyBz=0at the upper physical face. A potential field retains the padded-domain mean flux as a uniformk=0` field through the top. A nonzero-alpha field still requires a flux-balanced magnetogram and stops explicitly at a finite-height resonant mode rather than regularizing a singular solution silently.

For a nonzero-alpha LFFF, ‘lfff_flux_treatment='strict’keeps the default behavior and rejects a relative bottom-flux imbalance above1e-8. The opt-insubtract_meanmode first measures abs(sum(Bz))/sum(abs(Bz))on the cropped physical magnetogram. It subtracts the physical-core mean only when that value is no larger than lfff_max_flux_imbalance; more strongly imbalanced or nearly unipolar maps are rejected. The correction happens before FFT zero padding and never overwrites the input boundary file. Potential fields (lalpha=0`) are not modified by this option.

This PotentialField case deliberately requires Cartesian 3D, a uniform level-one mesh (refine_max_level=1), and cell-centered magnetic fields (stagger_grid=.false.) for both extrapolation methods. The physical magnetogram core must match the level-one horizontal cell centers exactly. Extra boundary ghost pixels are located and cropped by their coordinates; the solver does not assume a fixed ghost width and does not resample the data. Unsupported FFT sizes stop with their prime factorization and a suggested supported size.

The case uses the standard data-driven coronal normalization unit_length=1e9 cm, unit_temperature=1e6 K, and unit_numberdensity=1e9 cm^-3. AMRVAC derives the magnetic-field unit from these values. Density is uniformly one in code units, velocity is zero, the energy equation and gravity are disabled, and no GLM psi variable is stored.

The example contains a base amrvac.par next to mod_usr.t. The Python notebook can stage a user case from this directory and write a data-specific override file, typically data_driven_boundary.par, containing the generated boundary_filename and mesh extents. Run the staged case with both parameter files, for example:

mpirun -np 4 ./amrvac -i amrvac.par data_driven_boundary.par

The self-contained FFT backend has a small direct-DFT and mixed-radix round-trip test. The potential/LFFF vertical mode formulas have a separate analytic test. Run them with:

bash run_fft_core_test.sh

bash run_lfff_fft_test.sh