AI Inverse Design

Traditional materials search starts with a composition and looks up its properties. Inverse design flips this: you specify the properties you want, and MatCraft ranks materials from its database by how closely they match your targets.

How It Works

Define target properties: Set desired values for band gap, density, formation energy, or any other available property
Set importance weights: Assign relative importance to each property (0.0 to 1.0)
Choose constraints: Optionally require specific elements, crystal systems, or stability thresholds
Run the search: MatCraft scores all 205k+ materials against your target profile
Review ranked results: Materials are sorted by a weighted distance score, with the best matches at the top

Scoring Algorithm

The ranking uses a normalized weighted Euclidean distance:

score = sum(w_i * ((p_i - target_i) / range_i)^2)

Where w_i is the weight for property i, p_i is the material's property value, target_i is your target, and range_i is the property's range across the database for normalization.

Example Use Cases

Solar absorber: Target band gap 1.1-1.5 eV, low Ehull, density < 6 g/cm3
Wide-gap semiconductor: Target band gap 3.0-5.0 eV, cubic crystal system
Lightweight structural: Target high bulk modulus, density < 4 g/cm3, Ehull < 0.05 eV/atom

Using the Interface

Navigate to the Inverse Design tab from any materials page. The interface presents property sliders for each target value and weight. Results update in real-time as you adjust parameters.

API Access

python

import requests

response = requests.post("https://api.matcraft.ai/api/v1/materials/inverse-design", json={
    "targets": {
        "band_gap": {"value": 1.4, "weight": 1.0},
        "density": {"value": 3.0, "weight": 0.5},
        "e_above_hull": {"value": 0.0, "weight": 0.8}
    },
    "constraints": {
        "must_contain": ["Si"],
        "max_elements": 3
    },
    "limit": 50
})

candidates = response.json()["results"]

Limitations

Inverse design searches within the existing database — it does not generate novel compositions. For truly generative design, combine inverse design results with the optimization campaign workflow to explore nearby composition spaces around top candidates.