Constraining Chemical Simulation to Molecules That Can Actually Be Made

The central failure of most computational chemistry and AI-driven molecular generation is not imprecision but irrelevance: proposed molecules for which no synthetic route exists are not approximate answers — they are non-answers. The field has largely conflated theoretical chemical space (all thermodynamically stable arrangements of atoms) with accessible chemical space (molecules that can actually be synthesized given known reactions and available instrumentation). This conflation renders vast swaths of computational output scientifically meaningless.

The corrective requires constraining computational exploration to synthesizable molecules. Assembly theory provides a principled basis for this constraint by defining molecular identity not merely as a stable configuration but as an object that can be assembled through a sequence of known operations and produced in detectable quantities. When combined with a formally complete chemical programming language capable of expressing any known reaction and with physical robotic platforms that execute those programs, the result is a closed loop: simulation proposes, synthesis validates, and the space of exploration is bounded by physical reality rather than combinatorial fantasy.

This convergence mirrors what occurred in semiconductor design, where electronic design automation matured to the point that simulation replaced most physical prototyping. Chemistry is approaching an analogous threshold. The moment a chemistry simulator becomes trustworthy enough that its outputs reliably correspond to makeable molecules, the discovery cycle compresses from years to days. The rate of molecular discovery would not merely increase — it would undergo a Phase transition, fundamentally altering the economics and epistemology of chemical research.