Jeremy L. England (2013), The Journal of Chemical Physics 139(12):121923 (online 21 August 2013; CC BY 3.0); first posted as arXiv:1209.1179, v1 6 September 2012 (physics.bio-ph), the version read.
The brief’s citation (J. Chem. Phys. 139(12):121923, 2013) is correct.
Key takeaways
- A second law for macrostates (Eq. 6). If the dynamics obey detailed balance in a bath (no external driving), then for any two ensembles I and II defined by an observer’s classification of microstates, β⟨ΔQ⟩ ≥ −ln π(II→I) − ΔS_int. The heat released must pay for the internal entropy change and for how unlikely the return trip is.
- Replication is priced by speed and durability. For a replicator whose copies decay with lifetime τ (peptide-bond hydrolysis, RNA backbone hydrolysis) and which divides in time τ_div, the reverse probability is about the probability that a copy falls apart within τ_div. The heat bound grows with ln(τ/τ_div): a more durable or faster copy costs more heat.
- Worked estimates. E. coli releases 220 n_pep k_BT per division against an irreversibility term of about 75 n_pep, so less than three times the bound. The bound for a ligase-type RNA replicator (about 7 kcal/mol) sits near the measured reaction enthalpy (about 10 kcal/mol), and the same reaction in DNA would need about 16 kcal/mol, more than it releases. The author offers this as a reason RNA, not DNA, would come first.
- No physical “self”. “Self-replication is something that happens relative to an observer”: the count of copies exists only once a classification scheme assigns one to each microstate.
Standing in the record
Filed on 2026-10-02 at the owner’s request, as part of filling out the record’s coverage of dissipative structures and of the “life as dissipation” literature. No anthology topic holds it, and it carries no instruction for machine-learning practice.
NOTE-419 is the close reading of 2026-10-02, and it placed the work: Active. Its general result is a correct coarse-grained second law; the paper itself says it is closely related to the Landauer bound and holds for time-symmetrically driven systems too. The replication numbers depend on an upper bound for the reverse probability that is argued from biology (the most likely way back is one copy hydrolysing), not derived.
Where it bears on what the record holds:
- Landauer, Bennett and Still. The paper’s own comparison is with Landauer, whose paper the record holds unread (LIT-328). The record’s account of Landauer’s scope, THEORY-030, read from Bennett (LIT-360), says copying onto a blank register has no minimum cost. England’s bound does not contradict that. What it prices is not the copy but the finite-time irreversibility of making a durable copy quickly: as τ_div grows, the irreversibility term falls away. Still et al. (LIT-327) price the nonpredictive part of a driven system’s memory; England prices statistical irreversibility between macrostates. Both are second-law bounds over averages, and neither is per operation.
- The demarcation question (LIT-192). Nahas & Sachs report a dispute over whether teleological systems can be told apart from dissipative structures (NOTE-094). This paper takes the deflationary side: the replicating “self” is an observer’s coarse-graining, and the bound applies to any replicator with a reliable population model. It draws no line between a cell and a non-living replicator. The record’s other readings that make individuality observer- or model-relative are the Wilson & Barker survey (LIT-168) and Levin’s “Self” bounded by what it can measure and change (LIT-439); England’s coarse-graining is the statistical-mechanics version of that move.
- “What Lives?” (LIT-211). Its “Dissipative Self-Organizing Systems” cluster of definitions (NOTE-109) is the side of the definitional landscape this paper speaks for, but the paper defines no life.
Its sequel, Perunov, Marsland & England (2016), LIT-518, on “dissipative adaptation”, is filed beside it in this batch and drops self-replication altogether.