Getting started#
This section takes you from nothing installed to a real screened crystal structure on your own machine. Read it in order — it is written to be followed once, top to bottom.
What MatMelt needs from you#
Two things, and it is worth knowing both before you start.
A Python 3.10+ environment. The tool itself is small — a CLI, a config schema, an orchestrator and a state database. It installs in seconds and pulls in nothing heavy.
A scientific stack that you own. MatMelt does not bundle SMACT, PyXtal or MACE, and it never installs them behind your back. Each stage declares what it needs, and you point that stage at an interpreter where it is importable. This is deliberate — those packages are large, opinionated about their versions, and frequently already installed in the environment a researcher works in — but it does mean the first run involves real setup rather than one command. The tool and the science stack explains exactly what is needed and how the tool tells you what is missing.
How long this takes#
Step |
Time |
|---|---|
Install MatMelt |
under a minute |
Install the science stack, if you do not already have one |
5–20 minutes, mostly downloads |
Scaffold a config and get |
a few minutes |
Generate and screen your first structure |
under a minute of compute |
The compute really is that quick at this scale: two elements, one composition, 37 candidate structures generated in about ten seconds, and a single structure relaxed and screened in about fifteen. The funnel gets expensive when you widen the element set — which is what the remote backends exist for.
What you will have at the end#
A workspace directory containing a state database and one directory per
material, a screened structure with its relaxation verdict, and an
OPTIMADE-shaped record.json you can hand to another tool. From there, the
tutorials widen the funnel,
move stages onto other machines, and add the
optional higher-fidelity calculation stage.