Why template-directed replication and not just autocatalysis? Both multiply molecules. Both consume free energy. Both can be selected for in dissipative environments. The question of why life chose one over the other has been treated as historical contingency — a frozen accident of early chemistry.
This framework derives the answer from statistical physics. Dissipation-driven selection — the preferential survival of states that absorb and dissipate environmental free energy — creates a bias between the two strategies that is doubly exponential.
The argument: template-directed replication copies information with fidelity ε. Each round of replication produces offspring that are themselves capable of template-directed replication with the same fidelity. The population grows exponentially, and the fidelity of the exponentially growing population is itself exponentially maintained. This is doubly exponential amplification — the information content of the population grows as exp(exp(t)).
Simple autocatalysis, by contrast, amplifies the number of molecules but not the information they carry. Each copy is independent — errors don't compound, but neither does fidelity. The amplification is singly exponential.
In a competitive environment where free energy is limited, the doubly exponential strategy dominates the singly exponential one above a threshold of replication fidelity and resource availability. Below the threshold, neither strategy is stable. Above it, template-directed replication wins not gradually but categorically — the exponential separation between the two growth rates means that any finite advantage in fidelity translates to total dominance over any finite time horizon.
Life's defining feature — template-directed replication rather than mere self-catalysis — becomes not a contingent invention but a thermodynamic inevitability. Given sufficient fidelity and free energy, the universe selects for it. The question shifts from “why did life choose replication?” to “what set the fidelity threshold?”
Note (2026-11-11): checked against the full text of the paper (arXiv 2603.15230, a large-deviation argument for the selection of heredity). The framework (a bias toward histories with greater integrated dissipation), the comparison of simple autocatalysis with adaptive template-directed replication, the doubly exponential growth of the ratio of probabilities, and the thresholds for fidelity, kinetic efficiency and resource supply are in the paper. Four points differ. (1) The essay says the doubly exponential growth is in the information content of the population, which it derives from copying with fidelity epsilon (population grows exponentially, fidelity exponentially maintained, 'exp(exp(t))'). In the paper the autocatalytic history has an exponential increase in dissipation and the replicative history, through an effective adaptation rate alpha, a super-exponential one (the log of the dissipation grows like a constant times tau plus alpha over 2 times tau squared); it is the ratio of the probabilistic weights of the two histories that grows doubly exponentially, for large tau. (2) The essay says the dominance holds 'in a competitive environment where free energy is limited'; the model in the paper assumes a non-depleting resource supply (a continuous external flux keeps the fuel roughly constant) and says it neglects spatial heterogeneity and compartmentalization. (3) The essay says that 'any finite advantage in fidelity translates to total dominance over any finite time horizon'; the paper's statement is asymptotic ('for sufficiently large tau'), and it says that if the mutation rate is too high alpha becomes zero or negative and the heritable advantage is erased. (4) The essay says the result makes template-directed replication 'a thermodynamic inevitability'; the paper calls it a probabilistic bias, 'asymptotically dominant', conditional on thresholds, and proposes a falsifiable experiment (a convex curve in the logarithm of power dissipation against time) to test it, which the essay does not mention.