Reproducing the paper

The MANDALA paper demonstrates the framework on three materials and reports four controlled ablations and an inference-scaling benchmark. This section maps every result to the files and commands that produce it.

Result	Produced by	Section
ZnCu₂Sn(SeS)₂ Hamiltonian, correlation	`scripts/reproduce_examples/zncusnses.json`	Demonstration models
SiO₂ Hamiltonian, band energy, DOS	`scripts/reproduce_examples/sio2.json`	Demonstration models
Silicon H, D, S, band energy, bands, DOS, density	`scripts/reproduce_examples/silicon.json`	Demonstration models
Energy, spectral, envelope and aggregation ablations	`sweeps/paper_ablations`	Ablations and scaling
Inference scaling to 4096 atoms	`scripts/silicon_cif_scaling.py`	Ablations and scaling

Reference data

All operators were computed with OpenMX 3.9 (PBE, norm-conserving PBE19 pseudopotentials, non-spin-polarized, SCF threshold \(10^{-8}\) Hartree):

Dataset	Structures	Basis, grid, k-mesh, model cutoff
ZnCu₂Sn(SeS)₂	100 perturbed 32-atom cells	Zn10.0S-s2p2d2, Cu10.0S-s2p2d2, Sn7.0-s2p2d3f1, Se9.0-s4p3d3f2, S9.0-s4p3d3f2; 400 Ry; \(11^3\); 11 Å
Amorphous SiO₂	165 MD configurations of 150 atoms (100 used)	Si9.0-s2p2d1, O7.0-s2p2d1; 600 Ry; \(\Gamma\); 10 Å
Perturbed silicon	96 eight-atom diamond cells (80 train / 16 validation), \(\sigma=0.02\) Å	Si7.0-s2p2d1; 400 Ry; \(8^3\); 8 Å

Each structure is one directory containing HS.out and the OpenMX output file. The training launchers discover structures in these layouts:

ZnCuSnSeS/scale_1/<structure>/{HS.out, ZnCuSeS.out}     dataset_kind: ZnCuSnSeS
SiOx/<structure>/{HS.out, SiO2.out}                     dataset_kind: siox
perturbed_Si/scale_1/<structure>/{HS.out, Si.out}       dataset_kind: silicon_scales

Environment

Install MANDALA as in Installation with the CUDA stack. The paper models were trained in single precision on one NVIDIA GPU per model (B200 for the scaling benchmark). Training uses the deterministic settings of the launchers (PYTHONHASHSEED, CUBLAS_WORKSPACE_CONFIG=:4096:8, zero data-loader workers, fixed split and training seeds).

Workflow overview