MatForge combines surrogate models, active learning, and Pareto optimization to find optimal materials 100x faster than brute-force search. No GPU required.
Free to explore · No credit card required · 11 material domains
A complete platform combining physics-based evaluation, machine learning surrogates, and intelligent optimization.
NumPy-only MLP neural network with MC Dropout uncertainty. No GPU needed - runs anywhere.
Smart sampling with MaxUncertainty, Expected Improvement, and UCB acquisition functions.
NSGA-II multi-objective optimization with CMA-ES on the surrogate surface.
Water, battery, solar, CO2, catalyst, hydrogen, construction, bio, agri, electronics, textile.
Replace expensive simulations with surrogate predictions. Evaluate thousands in seconds.
Live campaign progress with 3D visualizations, Pareto plots, and convergence tracking.
Write a YAML material definition: parameters, objectives, constraints, and physics equations.
Start a campaign. The engine samples initial materials and trains a surrogate model.
CMA-ES finds optimal candidates on the surrogate. Active learning picks the most informative.
Pareto-optimal materials emerge. Export recipes, visualize trade-offs, iterate.
Each domain includes physics equations, YAML templates, and pre-configured optimization parameters.
PFOS rejection membranes
Next-gen energy storage
Perovskite photovoltaics
Carbon capture sorbents
Reaction optimization
H2 storage materials
Low-carbon concrete
Biocompatible scaffolds
Controlled-release fertilizers
Semiconductor materials
Responsive fabric composites
Heat-to-electricity conversion
High-Tc superconducting materials
Sustainable polymer design
Protective thin film coatings
High-performance ceramics
Evaluate 10,000 candidates where you could only test 50.
Equations from domain experts. Not just statistics.
No GPU, no cloud credentials, no Docker. pip install and go.
Write your own plugins. Bring your own evaluator.
name: PFOS Rejection Membrane
domain: water
parameters:
- name: pore_diameter
range: [0.5, 5.0]
unit: nm
- name: active_layer_thickness
range: [50.0, 500.0]
unit: nm
objectives:
- name: pfos_rejection
direction: maximize
equation: "water:pfos_rejection"
- name: permeability
direction: maximize
equation: "water:permeability"
active_learning:
initial_samples: 20
samples_per_round: 10
acquisition: max_uncertaintyJoin researchers using MatForge to discover novel materials for water, energy, construction, and beyond.