Reformix AI

The way we make materials is changing. For a century, new plastics have been found through slow cycles of mixing, curing, testing and starting over, a process measured in months and paid for in waste. Recent progress in simulation and machine learning suggests this is ending. Within a decade, most new materials will be designed on a computer first, and made in the lab only once we already know how they will behave.

Recyclability is the hardest test. Thermosets give us strength, chemical stability and heat resistance, but their permanent crosslinks mean they cannot be reprocessed or repaired. Vitrimers keep that strength and add the ability to rearrange, so a material can be reshaped, healed and recycled. To design one, you need to know how its network actually forms, atom by atom, and how that network will behave once it has.

Current methods fall short. Coarse-grained models discard the chemistry. Distance-based crosslinking forms a bond whenever two atoms happen to be close, ignoring geometry, steric hindrance and the real reaction path. And real crosslinking takes minutes to hours, far beyond what atomistic dynamics can reach. The result is networks that look plausible and predict poorly. Chemistry cannot be approximated by proximity. It has to be simulated.

We build the simulation methods that close this gap. We use biased reactive molecular dynamics to drive real reactions, such as epoxide ring opening, along their true mechanism at realistic temperatures. Then we grow the resulting networks and measure density, glass transition and mechanical response, all at full atomistic detail. Our first results are public as a preprint. We are working toward recyclable, self-healing polymers that are designed, not stumbled upon.