Nikolai Perunov, Robert A. Marsland III and Jeremy L. England (2016), Physical Review X 6(2):021036 (published 16 June 2016; CC BY 3.0); first posted as arXiv:1412.1875, v1 5 December 2014 (physics.bio-ph), the version read.
The brief’s citation (Phys. Rev. X 6(2):021036, 2016) is correct. The arXiv preprint lists the authors as “Nikolai Perunov, Robert Marsland, and Jeremy England”; Crossref has Perunov, Marsland, England.
Key takeaways
- An exact relation, rearranged. For two outcomes II and III of a system prepared in I and driven by a field λ(t), the log ratio of their forward probabilities equals the difference in internal entropy, plus the log ratio of their reversal probabilities, plus ΔΨ − ΔΦ. Ψ is the mean heat released (in k_BT) and Φ corrects for its fluctuations, so Ψ − Φ = −ln⟨e^{−βΔQ}⟩ (Eqs. 6–8). The authors call this “a generalization of the Helmholtz free energy” for driven, finite-time evolution.
- Dissipative adaptation, the interpretive claim. All else equal (same internal entropy, same reversal probability), the likelier outcome is the one reached by paths that absorbed more work from the drive and dissipated it more reliably. Such states have “dynamical response properties” tuned to the drive, as a resonant oscillator is tuned to its forcing frequency, and the paper calls them “better adapted”.
- No replication needed. “The adaptation we predict is expected to take place independent of whether or not there is anything in the system that can copy itself.” Self-replication is offered as one especially good way to keep absorbing work while changing shape. “Well-adapted” structures “that did not have parents” (sand dunes, snowflakes, hurricanes, active filament bundles) are proposed as test cases.
- What is shown. Two three-state and two-state hopping models with oscillating energies and barriers. In one, drift is reliably accompanied by dissipation, with r_max(2→1)/r(2→3) = exp(ΔΨ) (Eq. 12). In the other, fluctuations in entropy production slow the forward rate, ln r ≃ −Φ (Eq. 13).
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 the “life as dissipation” literature. No anthology topic holds it. It carries no instruction for machine-learning practice: its “learns” is in scare quotes, and it is about driven matter, not learning systems.
NOTE-416 is the close reading of 2026-10-02, and it placed the work: Active. It is the paper the phrase “dissipative adaptation” comes from, and it is cited far more widely than the evidence it gives. Its exact part is a rearrangement of England’s macrostate relation for a driven system. The “tendency towards adaptation” depends on holding the internal-entropy and reversal terms fixed. The paper shows that this can be done only in a contrived single-particle landscape, and argues that some of the ∼10²⁵ directions of a many-body phase space “should” behave the same way. The simulation of a driven chemical mixture is reported in one paragraph “rather than shown here”.
Where it bears on what the record holds:
- The predecessor. England 2013, LIT-515, supplies the macrostate relation (its Eq. 6) and the bound on replication heat that this paper’s Discussion invokes for Darwinian competition.
- Still et al. (LIT-327) and THEORY-026. Both papers study a system driven by a fixed external protocol with no feedback, and both read dissipation as a record of the drive. They use “adapted” in opposite directions. In THEORY-026 a system matched to its drive is one whose memory is predictive, and it wastes less work. Here a system matched to its drive is one that absorbs and dissipates more work, reliably. The two are not in contradiction. Still prices the dissipation of a given system over a protocol; this paper asks which macrostate a stochastic evolution is likely to reach. But the record should not merge the two senses, and a claim that “adapted systems dissipate more” or “adapted systems dissipate less” needs to say which one it means.
- The demarcation question (LIT-192, NOTE-094). The paper is the sharpest statement in the record of the view that there is no principled line between living adaptation and the self-organization of dissipative structures. It calls it “somewhat arbitrary how we have decided to draw a line around a whale”. Its stated aim is a notion of adaptation that does not need “to decide in advance on a physical definition of where one replicator ends, and another begins”. This is the claim Nahas & Sachs’s candle-flame dispute is about, made from the physics side. The paper does not engage the philosophical side (autopoiesis, closure, Deacon).
- Self-organization in the record. The primitive metabolic cycles of LIT-036 are self-organized states of a chemically driven system. This paper’s claim predicts that such states should be the ones that absorb and dissipate the chemical drive reliably. LIT-036 does not compute dissipation, so the prediction is untested there.
- “What Lives?” (LIT-211). This paper is a paradigm member of the “Dissipative Self-Organizing Systems” view of life (NOTE-109), though it defines adaptation, not life.
The paper cites the record’s neighbours on entropy-production principles (Martyushev 2010, as a contrast: maximising Ψ alone “is not predicted to correspond to a likely outcome”) and on fluctuation theorems (Crooks 1999, Jarzynski 1997, 2006; Hatano–Sasa 2001). Those are being filed in parallel in this batch.