In industrial practice, simulation models for new substance classes are often recalibrated against additional measurements – density, enthalpy, viscosity – before they can be trusted. Each new class repeats the cycle.
The Molecular AI Engine takes a different route: its parameters are built per chemical element, not per substance class. Five numbers were calibrated once, on ethane, butane and polyethylene. Torsion profiles, bond angles and crystal lattices are not entered – they emerge.
First internal results are encouraging: the model builds the polyethylene crystal cell on its own within 1–2% deviation, and reproduces the polypropylene chain helix to 0.3% without a single new number. Carbon and hydrogen are covered today; nitrogen and oxygen are in progress.
The method is at an early stage, and we say so openly. That is why we test it blind against real measurements together with industry partners – the only kind of evidence that counts.