3. Building alloys & supercells (SiGe)
The previous tutorials used tiny primitive cells. This one shows how LatticeMind builds larger and more complex structures for you — here, a random silicon-germanium alloy supercell — straight from a plain-language description.
3.1. The prompt
Do an SCF and band structure calculation for a 32-atom SiGe alloy with a
Ge fraction of 0.5, and place the Ge atoms randomly.
3.2. What LatticeMind does
From that one sentence, LatticeMind’s structure builder:
starts from the diamond-cubic silicon lattice and constructs a 2×2×2 supercell — 32 atoms in a cubic cell of side ≈ 10.86 Å;
randomly assigns half the sites to germanium (Ge fraction 0.5, so 16 Si and 16 Ge);
defines both species with their PBE pseudopotentials; and
verifies the structure is physically sane (composition, site count, bond lengths) before any calculation runs.
The generated cell carries both elements, for example:
domain.latvec = [...
10.86 0.0 0.0;
0.0 10.86 0.0;
0.0 0.0 10.86;
];
element(1).species = 'Si';
element(1).path = './pseudopotentials/Si_PBE_TM_DZP.mat';
element(2).species = 'Ge';
element(2).path = './pseudopotentials/Ge_PBE_TM_DZP.mat';
LatticeMind then plans the SCF and band-structure steps on this 32-atom cell, using a k-point grid appropriate for the larger cell (a supercell needs fewer k-points than a primitive cell).
3.3. Varying the alloy
Everything about the structure is controllable in words. Try:
Make it a 64-atom cell with Ge fraction 0.25.
Use an ordered SiGe arrangement instead of random.
Add 1% biaxial strain in the xy-plane.
LatticeMind rebuilds the structure and re-plans accordingly.
Tip
The same builder handles supercells, alloys, point defects, surfaces, and two-probe device geometries. If you can describe the structure, LatticeMind can usually build it — and it will tell you when a request is ambiguous and ask for the missing detail rather than guessing.
3.4. What you learned
LatticeMind builds non-trivial structures — supercells and random alloys — from a description, and validates them before running.
Structural parameters (size, composition, ordering, strain) are all adjustable through natural language.
That completes the basic tutorials. Continue to Advanced tutorials for remote/HPC execution and the experimental NanoDCAL backend.