The Digital Twin: 262,000 Simulations of Third Matter

After months of computational work, we have completed ORACLE — our global sensitivity analysis of the Third Matter system. The numbers tell a clear story.

What we did

We sampled the 16-dimensional parameter space of our G3c membrane model using Sobol sequences, running 262,000 stochastic (Gillespie) simulations. A Random Forest classifier (AUC = 0.982) then mapped the boundary between “alive” and “dead” states.

What we found

The system is transport-limited, not reaction-limited. The catalytic rate constant k_cat is completely decoupled — varying it over 7 orders of magnitude (10⁻⁸ to 10⁻¹) changes the steady-state formate concentration by less than 0.01 mM. The mackinawite layer thickness L_mack is the only real control knob:

• 20 nm → 0.9 mM formate
• 50 nm → 3.2 mM
• 100 nm → 6.5 mM

The three critical parameters are formate degradation rate (kd_A, importance 0.36), formate production rate (k1, 0.13), and membrane degradation rate (kd_m, 0.10). Six parameters have no detectable threshold at all.

What this means

The membrane is the metabolism. Third Matter doesn’t need a fast catalyst — it needs a good membrane. This simplifies the experimental design considerably: we can focus on mackinawite layer thickness and membrane stability rather than chasing catalytic efficiency.

The digital twin is complete. Now we build the real thing.