Skip to content

PhaseForge

Optional dependency

Sqs2tdb requires the calphad extra, plus the sqs2tdb binary (part of ATAT) on PATH, with its SQS database configured via ~/.atat.rc.

uv add "materialsframework[calphad]"

pip install "materialsframework[calphad]"

Fits a CALPHAD thermodynamic database (.tdb) solution model directly from SQS energies computed with a MaterialsFramework calculator, bridging MLIP energetics into standard CALPHAD-format phase-stability data.

Overview

CALPHAD phase diagrams and stability maps (see StabilityMap) are built from Gibbs-energy solution models whose parameters are traditionally fit to a mix of experimental data and DFT formation energies. Sqs2tdb automates a materials-informatics shortcut: it pulls Special Quasirandom Structures from ATAT's built-in SQS database for each requested lattice and composition (every lattice in that database if lattices is left as None), evaluates their formation energies with a BaseCalculator (or, for the "LIQUID" lattice, by equilibrating with NPT-then-NVT molecular dynamics via a BaseMDCalculator, on a supercell scaled by md_scaling_matrix), and hands those energies to ATAT's sqs2tdb script to regress a substitutional solution model, written out as a standard .tdb file readable by pycalphad or Thermo-Calc. Structures are converted from ATAT's own format straight to POSCAR with pymatgen.io.atat.Mcsqs, so no VASP installation is needed even though the underlying script is ATAT's VASP-oriented sqs2tdb.

Theory

For each lattice being fit (e.g. "BCC_A2", "FCC_A1", "LIQUID"), the molar Gibbs energy of the substitutional solution phase follows the standard CALPHAD Redlich-Kister-Muggianu form:

\[ G_m = \sum_i x_i\,{}^{0}G_i + RT \sum_i x_i \ln x_i + \sum_{i<j} x_i x_j \sum_{v} {}^{v}L_{ij}\,(x_i - x_j)^v \]

where \(x_i\) is the mole fraction of component \(i\), \({}^{0}G_i\) is the end-member reference lattice stability, the middle term is the ideal configurational entropy of mixing, and \({}^{v}L_{ij}\) are Redlich-Kister interaction parameters of order \(v\) between components \(i\) and \(j\). fit()'s terms argument controls which of these are regressed for each lattice, via (order, level) pairs: order=1 fits an end-member-like linear-in-composition term, order=2 fits a binary interaction \({}^{v}L_{ij}\) at Redlich-Kister degree level, and order=3 extends this to ternary interactions. terms can be a single string or list of pairs applied to every lattice, or a dict keyed by lattice name for per-lattice control, with any lattice left out of the dict falling back to the built-in default. The regression uses the SQS formation energies, relaxed at 0 K or time-averaged over the last 20% of an MD trajectory for "LIQUID", as the enthalpic data fit against these composition-dependent terms over the temperature window [t_min, t_max]; sro=True additionally accounts for short-range order beyond the ideal-mixing entropy term.

References