Contribution
Before this paper, fluctuation relations had been used for the thermodynamics of copying informational polymers (Andrieux & Gaspard 2008, its ref. [13]), where what counts as a copy is fixed by the chemistry. This paper states the relation for arbitrary coarse-grained ensembles and uses it to bound the heat of self-replication of a whole organism, with the replicating “self” supplied by an observer’s classification. The bound links heat to growth rate, internal entropy change and durability. It gives the first quantitative estimate of how close a bacterium runs to that bound.
Key insight
Heat released in a transition between macrostates must pay for how unlikely the transition is to run backwards. A replicator whose copies are durable and made quickly runs a transition that is very unlikely to reverse, so it must release a lot of heat. The cost is set by irreversibility on the timescale of one generation, not by the information in the copy.
Assumptions
- Dynamics. Stochastic dynamics in a heat bath at fixed temperature, with a transition matrix obeying detailed balance, π(j→i)/π(i→j) = exp(−βΔQ_{i→j}) (Eq. 2). “By stipulation, there are no external driving forces,” and the motion is diffusive, “lack[ing] any sense of momentum.” The text adds that the same form holds for a time-symmetric drive, by Crooks’s relation, without deriving it here.
- Ensembles. I and II are probability densities over microstates fixed by an experimental preparation and, in principle, by “repeated consultations of a microbiologist.” The paper does not compute them.
- The reverse probability. The bound needs an upper bound on π(II→I). The paper argues one from biology. The most likely route back is one cell dividing slightly early and a daughter hydrolysing completely, which is more likely than a growing cell pausing for a whole generation (Fig. 1). Hence ln π(II→I) ≤ 2 ln p_hyd (Eq. 7), with ln p_hyd ≃ n_pep ln[τ_div/(n_pep τ_hyd)] from Poisson statistics. This is a plausibility argument about which path dominates, not a bound proved over all paths.
- Internal entropy. −ΔS_int ≤ 10 n_pep is set “arbitrarily” and “generous[ly]” from two estimates: O₂ to CO₂ (+6 per carbon) and amino acid confinement (about −12 per residue).
Key results
- Eq. (6), the general bound. From ⟨exp(−βΔQ − ln π(II→I) − ln p(j|II) + ln p(i|I))⟩ = 1 (Eq. 5) and Jensen’s inequality (the text says “e^x ≥ 1 + x”), β⟨ΔQ⟩ + ln π(II→I) + ΔS_int ≥ 0. Since π ≤ 1 it implies ⟨ΔS_tot⟩ ≥ 0. I re-derived it from Eqs. (2)–(5); given the definitions it is correct. The paper calls it “simply a precise statement of the Second Law” and “closely related to the well-known Landauer bound”.
- Eq. (8), the replication bound. β⟨Q⟩ ≥ 2n_pep ln[(n_pep τ_hyd)/τ_div] − ΔS_int.
- E. coli. With n_pep = 1.6×10⁹, τ_div = 20 min and τ_hyd = 600 years, 2n_pep ln[(n_pep τ_hyd)/τ_div] ≈ 75 n_pep ≈ 1.2×10¹¹. I recomputed this: n_pep τ_hyd/τ_div ≈ 2.5×10¹⁶, whose log is 37.8, so 75.6 n_pep. The measured heat is 220 n_pep (Rothbaum & Stone 1961). The ratio 220/75 ≈ 2.9 is the paper’s “less than three times as large as the absolute physical lower bound”.
- RNA and DNA. For a ligase-type self-replicating RNA (Lincoln & Joyce 2009) with a 1-hour doubling time and a 4-year half-life, ⟨Q⟩ ≥ RT ln[(4 years)/(1 hour)] = “7 kcal mol⁻¹”, against a reaction enthalpy near 10 kcal/mol. For DNA (half-life 3×10⁷ years), “16 kcal mol⁻¹”, which exceeds the enthalpy and is “prohibited thermodynamically”. I recomputed them: ln(35,064) = 10.5, so 6.2–6.5 kcal/mol at 298–310 K; ln(2.6×10¹¹) = 26.3, so 15.6–16.2 kcal/mol. The DNA figure checks; the RNA figure is rounded up by about 0.5–0.8 kcal/mol, which does not change the conclusion.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | For detailed-balance dynamics, heat released in a macrostate transition is bounded below by the log-odds against reversal minus the internal entropy change | strong | Eqs. (2)–(6); a correct coarse-grained second law |
| C2 | The same bound holds for time-symmetrically driven systems | moderate | asserted via Crooks [3], not derived here |
| C3 | Replication heat is bounded by 2n ln(n τ_hyd/τ_div) − ΔS_int | moderate | Eq. (8); depends on the argued bound Eq. (7) on the reverse probability |
| C4 | E. coli releases less than three times the minimum heat its growth rate, entropy change and durability allow | moderate | the arithmetic checks (220/75.6); the 75 n_pep rests on C3, and the paper says the reverse probability is deliberately overestimated, which would raise the bound and tighten the ratio |
| C5 | A DNA replicator of the RNA’s speed would be thermodynamically forbidden, favouring RNA first | weak | one comparison with assumed enthalpy and half-lives; the linear-in-ℓ extrapolation is stated, not computed |
| C6 | ”Self-replication” exists only relative to an observer’s classification of microstates | assertion | a framing argument (pp. 1, 4), not a result; the bound needs some coarse-graining but works for any one |
Concepts
- Coarse-grained ensemble. A probability density over microstates fixed by a macroscopic preparation or classification, I or II. No order parameter is needed.
- Reverse probability π(II→I). The probability of returning to ensemble I within the same time, starting from II.
- Durability. The decay time of the copy (τ_hyd for peptide bonds, the RNA or DNA backbone half-life). It enters the bound through ln(τ/τ_div).
Connections
- Andrieux & Gaspard (2008), ref. [13]. The paper’s own lineage for the thermodynamics of copying, where the information content of a template could “more easily be taken for granted”. Not held in the record.
- Crooks (1999), ref. [3]. The fluctuation relation the general claim rests on. It is being filed in parallel in this batch with the record’s other fluctuation-theorem work.
- Landauer, LIT-328 (Deferred). Cited by the paper as closely related. The record’s account of what Landauer prices is THEORY-030, from Bennett (LIT-360): only logically irreversible steps, and copying onto a blank register costs nothing. England’s bound is consistent with that. Its cost comes from making a durable copy in finite time, and it falls away as τ_div grows, as Bennett’s reversible-copying limit requires.
- Still et al., LIT-327 (NOTE-295). Another second-law bound read as an information cost, and THEORY-026’s source. Still bounds work dissipated by a system with fixed kernel, driven without feedback, by its nonpredictive memory. England’s system is undriven and his quantity is macrostate irreversibility. The two are not in tension: neither is per operation, and they price different things.
- Perunov, Marsland & England (2016), LIT-518. The sequel, which starts from the same macrostate relation, adds a drive and drops replication.
Bearing on the record
- THEORY-030 (Proposed): consistent, and a useful check. The paper says its bound is “closely related to the well-known Landauer bound”. On the record’s reading of Landauer’s scope, the heat England prices is not a cost of copying information. It is the cost of irreversibility at a finite rate, and it vanishes in the slow limit. A reader who took England as evidence that replication has an intrinsic per-bit cost would be making the mistake THEORY-030’s “What this does not say” warns against.
- THEORY-026 (Proposed): no direct bearing. England’s system is undriven and has no memory variable.
- LIT-192 and the demarcation question. The paper is explicitly deflationary: it supplies no physical mark of a living replicator, and locates “self” in the observer’s coarse-graining (C6).
- No ML instruction. Nothing here belongs in the anthology.
- No new THEORY is indicated from this paper alone.
Limitations
- The reverse-probability bound (Eq. 7) is a biological argument about which path dominates. The paper claims it as an upper bound (“we can claim that”), but no proof covers all paths back to I.
- The internal entropy bound is set “arbitrarily”.
- One organism and one RNA system are worked. The claim that the approach “applies equally” wherever a stochastic population model exists is not tested.
- The driven case is asserted by citation only.
Open questions
- A computed, not argued, bound on π(II→I) for a real replicator, which would turn C4 into a measurement.
- Whether the factor of three for E. coli is set by adaptability, as the author suggests (“not perfectly optimized for any given” environment), or by the deliberately loose reverse probability.
Corrections
- none to a seeded skim (there was no seed)
- An inverted logarithm. The text gives “2n_pep ln[τ_div/(n_pep τ_hyd)] = 1.2×10¹¹ ≃ 75n_pep”. That logarithm is negative (about −37.8 per bond). The positive value belongs to ln[(n_pep τ_hyd)/τ_div], as in Eq. (8). This is a sign slip in the prose, not in the bound.
- The RNA figure is rounded up. RT ln(4 years/1 hour) is 6.2–6.5 kcal/mol at 298–310 K, not 7.