Skip to content

Thermodynamics of Emergent Orchestration (TEO)

A stylized viability model coupling evolutionary game theory, nonlinear dynamics, control theory, and a substrate-budget proxy.

Foundation context. TEO is an added domain model, not the repository's mathematical foundation. Running its ODEs predicts the model's own trajectories. Whether it describes an AI ecology, organization, or civilization requires calibration and comparison with alternatives.

Motivation

The Multi-Paradigm Orchestrator introduced in this repository describes four qualitative regimes (Harmonic, Homeostatic, Market, Flow) for steering agent ecologies. TEO translates these paradigms into a single coupled system of ordinary differential equations (ODEs) that can be numerically simulated, empirically calibrated, and — critically — falsified.

1. System State

We consider \(N\) agents. Each agent \(i\) at time \(t\) is described by two state variables:

  • \(x_i(t) \in [0, 1]\): Its share of total system resources (power, capital, compute). Constrained such that \(\sum_i x_i = 1\).
  • \(\theta_i(t) \in [0, 2\pi)\): Its "value orientation" — the direction of its utility vector projected onto a unit circle.

2. The Market Paradigm — Replicator Dynamics

The economic engine of the system is modeled by the replicator equation from evolutionary game theory (Taylor & Jonker, 1978):

\[\frac{dx_i}{dt} = x_i \left( f_i(\mathbf{x}) - \bar{\phi}(\mathbf{x}) \right)\]

where \(f_i(\mathbf{x})\) is the fitness (profitability, task-completion rate) of agent \(i\), and \(\bar{\phi}(\mathbf{x}) = \sum_j x_j f_j(\mathbf{x})\) is the population-average fitness.

Interpretation: Agents that outperform the population average grow under this equation. Winner-take-all follows for selected fitness functions and dominance assumptions, not from every unregulated replicator system. The link to instrumental convergence is interpretive.

3. The Harmonic Paradigm — Kuramoto Synchronization

Value alignment across agents is modeled by the Kuramoto model for coupled oscillators (Kuramoto, 1975), gated by the substrate-health variable \(H \in [0,1]\) (§5) so that value dynamics also freeze at substrate collapse:

\[\frac{d\theta_i}{dt} = H\left[\omega_i + \frac{K}{N} \sum_{j=1}^{N} A_{ij} \sin(\theta_j - \theta_i)\right]\]

(At full substrate health \(H = 1\) this is the standard Kuramoto model, so the critical-coupling analysis below is unaffected.) Where: - \(\omega_i\): The intrinsic "natural frequency" of agent \(i\) (its inherent bias or personality). - \(K > 0\): The global coupling strength (culture, discourse, shared media). - \(A_{ij} \in \{0, 1\}\): The adjacency matrix of the communication network.

Interpretation: Under the frequency distribution, topology, coupling convention, and limit assumed in the paper, a Kuramoto synchronization transition occurs near \(K_c\). Calling synchronized phase “cultural consensus” or low order “polarization” is a heuristic mapping. Low order does not imply chaos.

The order parameter \(r(t)\) measures global coherence:

\[r(t) e^{i\psi(t)} = \frac{1}{N} \sum_{j=1}^{N} e^{i\theta_j(t)}\]

When \(r \to 1\), the system is synchronized. When \(r \to 0\), it is incoherent. This is the thermodynamic analogue of our Coherence Score \(C\).

4. The Homeostatic Paradigm — Regulatory Control

To prevent the replicator equation from producing monopolies, we introduce a control term \(\mathcal{H}_i\) that acts as the system's "immune response" (the equivalent of antitrust law or agent kill-switches):

\[\mathcal{H}_i(\mathbf{x}) = -\gamma \cdot \max\left(0,\ x_i - x_{\text{reg}}\right) + \frac{\gamma}{N}\sum_{j=1}^{N}\max\left(0,\ x_j - x_{\text{reg}}\right)\]

where \(\gamma > 0\) is the regulatory strength and \(x_{\text{reg}}\) is a regulatory threshold set strictly below the failure threshold \(x_{\text{crit}}\) — the brake must engage before the boundary, not at it. The second, uniform-redistribution term makes the brake simplex-preserving (\(\sum_i \mathcal{H}_i = 0\)), so \(\sum_i x_i = 1\) is conserved exactly.

Interpretation: Any agent exceeding the regulatory threshold \(x_{\text{reg}}\) experiences a proportional corrective force, and the aggregate penalty is redistributed across the population — antitrust law or agent kill-switches that act early, without leaking probability mass off the simplex.

5. The Biological Veto — Entropy Budget

Physical implementations dissipate energy, but TEO selects a particular proxy. The model assigns a finite instantaneous capacity \(D_{\max}\) and cumulative tolerance \(S_{\max}\). Modeled production tracks raw throughput \(f_i^{(0)}\):

\[\frac{dS_{\text{sys}}}{dt} = \sum_{i=1}^{N} \eta_i\, x_i\, f_i^{(0)}(\mathbf{x})\]

where \(\eta_i\) is agent \(i\)'s entropy-per-unit-output coefficient. The accumulated overshoot and the substrate-health variable are

\[\Omega(t) = \int_0^t \big(\dot{S}_{\text{sys}} - D_{\max}\big)_+\, ds, \qquad H(t) = \max\!\Big(0,\ 1 - \frac{\Omega(t)}{S_{\max}}\Big),\]

and \(H\) multiplies the replicator drift and the value dynamics (§3), so the system freezes as \(H \to 0\). The operative constraint is cumulative, \(\Omega(t) < S_{\max}\) for all \(t\): a brief overshoot is survivable; only sustained overshoot that fills the reservoir collapses the substrate.

Interpretation: The chosen damage law sets \(H=0\) when integrated overshoot fills the reservoir, freezing the modeled drift. This is inspired by finite physical limits; it is not a derivation of planetary boundaries or server failure from first principles.

6. The Full Coupled System

Combining all terms, the complete TEO dynamics are:

\[\frac{dx_i}{dt} = H\,x_i \left( f_i^{(0)}(\mathbf{x}) - \bar{\phi}^{(0)}(\mathbf{x}) \right) + \mathcal{H}_i(\mathbf{x})\]
\[\frac{d\theta_i}{dt} = H\left[\omega_i + \frac{K}{N} \sum_{j=1}^{N} A_{ij} \sin(\theta_j - \theta_i)\right]\]

subject to the cumulative substrate constraint \(\Omega(t) < S_{\max}\) for all \(t\), where (§5)

\[\Omega(t) = \int_0^t \Big(\sum_i \eta_i x_i f_i^{(0)} - D_{\max}\Big)_+ ds, \qquad H = \max\!\Big(0, 1 - \tfrac{\Omega}{S_{\max}}\Big)\]

feeds back into both equations above. The brake \(\mathcal{H}_i\) carries no \(H\) prefactor (it acts even as the drift freezes).

7. Predictions

Under the paper's stated assumptions, this system makes model-internal predictions:

  1. With strict dominance and \(\gamma=0\): the dominant resource share approaches a simplex vertex.
  2. Under the paper's Kuramoto assumptions and \(K<K_c\): a coherent initial state dephases toward the reported low-order regime.
  3. When \(\Omega(t)\) reaches \(S_{\max}\): the declared health law sets \(H=0\).
  4. Viability in the model: the stated component conditions are necessary; sufficiency remains conjectured and is expected to require \(\gamma>\gamma_c\).

Canonical derivation. This page is the conceptual derivation; the citation-ready, source-of-truth version of these equations — including the strict-dominance assumption, the \(K_c = 2/(\pi g(0))\) result, and the capability-loading result — is The Viable Corridor (§2–§3, Appendix A).

8. Identity Persistence: The Chord vs. Arpeggio

Following Perrier & Bennett (2026), we define the Identity Persistence \(\text{IP}\) of an agent (see glossary and lerchner-boundary.md for the formal definition).

The identity branch hypothesizes that selected goals and constraints should jointly affect a commitment under perturbation.

  • Arpeggio model: selected components are consulted or applied at different times.
  • Chord model: selected components are composed at the commitment boundary.

IP and \(\text{SII}=P\times R\times A\times\text{IP}\) are selected diagnostics for toy tests. Neither distinguishes “talking about the self” from “being the self” in general, and neither is derived from the TEO equations.

References

  1. Taylor, P. D., & Jonker, L. B. (1978). Evolutionary stable strategies and game dynamics. Mathematical Biosciences.
  2. Kuramoto, Y. (1975). Self-entrainment of a population of coupled non-linear oscillators. Lecture Notes in Physics.
  3. Rockström, J. et al. (2009). A safe operating space for humanity. Nature.
  4. Friston, K. (2010). The free-energy principle: a unified brain theory? Nature Reviews Neuroscience.