Solutions · Metals & alloys

Alloys: HEA screening, CALPHAD sections and first-principles validation

The public mirror of the Alloys studio. Density and cost are computed, yield strength is an empirical prediction, and CALPHAD states which boundary source actually ran — real pycalphad, an in-house regular-solution model, or an explicitly labelled fallback.

Jobs to be done

What alloys teams come here to do

Each job maps to a module that exists — the links go straight into the platform.

Featurize the alloy space — VEC, atomic-size mismatch, Miedema mixing enthalpy and the Omega parameter, with a solid-solution vs intermetallic risk verdict (/studio/alloys).
Read phase behaviour — Ternary and quaternary CALPHAD sections plus Scheil solidification — real pycalphad when the database is available, an honestly labelled fallback otherwise (/viz).
Screen with potentials — Relax and compare structures with the machine-learned potentials installed on this deployment (/studio/mlip).
Escalate to real compute — A real ASE EMT relaxation in-app, then Quantum ESPRESSO or VASP jobs on the Slurm cluster (/studio/dft).
Design under a cost budget — Search compositions toward target properties with $/kg enforced as a constraint (/studio/optimization).

Modules

The real pieces behind this workflow

No roadmap items and no imagined features: every card opens a shipped module route.

See all modules

Alloy studio

Metals & HEA Alloys

Miedema features, phase and sigma-risk verdicts, mechanical estimates and cost-aware design.

Phase diagrams

Advanced Viz

CALPHAD isothermal sections, pseudoternary slices and Scheil solidification for candidate systems.

Potentials

MLIP (MACE, CHGNet, M3GNet, EMT)

Registry-driven relaxation and MD; the page never assumes a potential is installed.

First principles

DFT & HPC

Real EMT relaxations in-app; QE/VASP jobs queued to Slurm with the engine card naming the path.

Optimization

Cost-constrained search

Hill-climb or evolutionary search toward targets under a cost budget and compositional bounds.

Evidence & limits

What the outputs can honestly support

Screening-grade where true; engine-labelled everywhere.

Which number came from which engine.Density and cost are COMPUTED; yield strength is an empirical rule labelled PREDICTED; phase and MLIP results are screening estimates. CALPHAD reports its boundary source: real pycalphad with a vendored TDB, the in-house regular-solution model, or an explicitly labelled DEMO result — boundaries are never fabricated. The in-app DFT composition path is a physics heuristic, not ab initio; the engine card names the path that produced the numbers.

Alloys domain pack →CALPHAD phase diagrams →DFT & quantum →Machine-learned potentials →

FAQ

Questions before you start

Including the limitations we would rather state now.

Is the phase prediction a real CALPHAD solve?

When a vendored TDB and pycalphad are available it is; otherwise the page falls back to an in-house regular-solution model or an explicitly labelled DEMO result. The response states which boundary source ran (client/src/pages/docs/content.jsx, visualization/calphad and domain-packs/alloys).

Are yield-strength numbers reliable?

They are an empirical estimate labelled PREDICTED, while density and cost are COMPUTED. Treat mechanical values as screening inputs and validate on samples.

Can Matflow run DFT?

The in-app composition path is a physics heuristic. A real ASE EMT relaxation runs for parameterized metals, and the DFT page submits Quantum ESPRESSO or VASP jobs to a Slurm cluster; the engine card says which path produced the numbers (docs/quantum/dft).

Do I need my own thermodynamic database?

CALPHAD ships with vendored TDBs for common systems (Fe-Ni-Cr, Co-Cr-Ni, Al-Ni-Ti, Ti-Al-V, Cu-Ni-Zn and HEA subsystems). Without one, results are explicitly labelled DEMO rather than guessed (docs/visualization/calphad).

Can I constrain cost?

Yes — design_alloy hill-climbs toward targets under a cost budget, and optimization supports per-feature bounds and compositional constraints.

Screen the alloy space, then spend compute on the winners

Featurize a composition, read its phase verdict, and send only the promising systems to MLIP relaxation or a real DFT job — free tier, no card.