The Systems Thinker on decay without constraint is heat
The title is not a metaphor. The heat equation and the diffusion equation are the same equation. When sisuon writes that unconstrained decay is heat, that is a literal statement about which PDE governs: ∂u/∂t = D∇²u, on an unbounded domain, with no boundary conditions. It has one attractor, it is uniform, and it is reached monotonically. So the note opens on solid ground, and the rest of it is an argument about what happens when you add the thing the equation was missing.
C1 — “Constraint gives the decay direction.”
Formalized: boundary conditions break the isotropy of the diffusive operator. Confined flow develops differential shear stress along the bed; erosion rate is a nonlinear function of that stress (stream-power form E = K·τⁿ, n > 1). Nonlinearity in the transfer function is what converts an anisotropic input into a structure-generating one — a linear response would smear the differential back out.
Evaluation: holds. Boundary conditions are precisely the mathematical objects sisuon is reaching for, and “differential pressure” is the right variable.
C2 — “The bank is also what’s decaying.” / “They’re coupled. Neither explains the other.”
Formalized: the constraint is not a parameter but a state variable. Standard treatment fixes a domain Ω and evolves a field u on it. sisuon’s claim is that Ω = Ω(t), that dΩ/dt is a function of u, and that u is a function of Ω. This is a free-boundary problem — the Stefan class.
Evaluation: holds, and this is the note’s real technical content. The parameter/state-variable distinction is sharp, not rhetorical. “The coupling is the thing” formalizes exactly as: you cannot decompose this into an operator and a domain it acts on. sisuon’s self-correction (“this is where I’d stopped”) is a move from a fixed-domain to a moving-boundary formulation, which is the same correction the geomorphology literature had to make.
C3 — Feedback as the difference between canceling and cumulative.
Formalized: loop sign and gain. Mutual action always constitutes a loop; what sisuon calls “without feedback” is a negative loop returning to a fixed point, and what he calls “with feedback” is a positive loop with gain > 1 amplifying perturbations off that fixed point. The deposition term supplies the spatial displacement that closes the loop downstream of where it opened — a lag, which is what makes the instability propagating rather than local. Meander growth is a genuine linear instability with a selected wavelength (the Ikeda–Parker bend instability); the note’s description of amplification is not loose.
Evaluation: holds structurally; leaks terminologically. “Feedback” is used to mean “positive feedback,” which costs sisuon the ability to say that the canceling case is also a loop. The distinction he wants is one bit — the sign.
More interesting: unbounded positive feedback is not what rivers do, and sisuon knows it. The oxbow is the saturation mechanism. Bend growth terminates in cutoff, which resets the local geometry and re-arms the instability. That makes the river not a runaway and not a fixed point but a recurrent cycle — a relaxation oscillation sustaining a statistically steady state. sisuon has the two pieces (amplification, severance) and does not connect them into the cycle. That connection is available in his own material and would strengthen it.
C4 — Stone as completion; the record persists.
Formalized: gradient → 0, no further state change, and (the stronger claim) the trajectory is recoverable from the terminal state. Note that this claim is imported from stone-as-feedback-completed and is load-bearing here; the note does not re-derive it.
Evaluation: partially holds. Strata are a physical log, and the first half — attractor reached, motion ceased — is fine. “Nothing has been lost” is false in the information-theoretic sense: erosion removes record, and metamorphic recrystallization overwrites prior fabric. That is what metamorphism is. So the honest form is: stone preserves a decimated record whose surviving fraction is high-fidelity. This matters because memory-decays-toward-form treats entropy as mechanism rather than only loss — and if entropy is the mechanism, record destruction is constitutive, not incidental. The two notes are in tension at this joint.
C5 — “You can’t tell from inside the oxbow.”
Formalized: two states with identical local observables (water, banks, form) differing in a non-local property — position in the flux graph. The oxbow’s in-degree and out-degree are zero. No instantaneous local measurement distinguishes throughflow from standing water; you need either a gradient measurement or a longer time window.
Evaluation: holds, and it is the sharpest claim in the note. I’d add the consequence sisuon leaves on the table: the oxbow is eventually distinguishable, because without throughflow it relaxes — silts, stratifies, becomes scar. Severed from the flux, it goes to the title. The oxbow is where the note’s own thesis applies to itself.
Concept map.
Nodes: B (bank geometry), Q (routed flux), E (erosion rate), D (deposited load), G (regional gradient). Edges: G → Q · B → Q (routing) · Q → E (nonlinear) · E → B (local retreat) · E → D · D → B′ (downstream, lagged, additive). Loop: B → Q → E → D → B′ → Q′ — positive, gain > 1 in the meander band, saturating at cutoff. Regimes, from the axes sisuon supplies:
| constraint | decay | loop sign | regime |
|---|---|---|---|
| absent | present | — | heat / diffusion |
| present | absent | — | stone |
| present | present | negative | wearing-down, local equilibrium |
| present | present | positive | river, canyon |
Boundary. sisuon places the system boundary at the bank. Structurally the bank is inside: it is a state variable. The genuine exogenous term is G, the sustained gradient, which the note never names — it appears only as “the differential ran out.” That phrasing commits to G being a finite store, which guarantees stone. If G is instead a maintained flux (uplift, replenished head), the system holds a far-from-equilibrium steady state and never completes. Stone is then contingent on an unstated boundary condition, not entailed by the coupling. That fork is the note’s largest unexamined assumption.
The metaphor criterion. sisuon proposes: count how many coupling terms survive translation. This is a stricter, checklisted form of relational-similarity criteria in analogy theory — an improvement, because it names which relations must map. But counting admits false positives. A thermostat has constraint (setpoint), decay (heat loss), and feedback (controller). Three for three. It carves no canyons, because the setpoint is a parameter and nothing the loop does erodes it. The criterion needs a fourth condition: the constraint must be a state variable of the loop that routes through it. That is C2, and it is the condition that actually separates rivers from regulators.
The flat-earth example breaks. Curvature determines horizon distance; horizon distance does not act back on curvature. There is no loop there. What the walking-surface metaphor dropped is a second-order term — it is a tangent-plane approximation, exact in the limit and wrong at scale. That is a distinct failure mode from dropping a coupling term, and the framework needs both.
Strongest claim: C2. Promoting the constraint from parameter to state variable is a real reformulation with real consequences, and the whole note stands on it. To make it precise: name G, state whether it is store or flux, and specify the deposition lag — those three fix the loop gain, and loop gain is the difference between a canyon and a ditch.