Kramers-Wannier Duality: When Misalignment Is Order

The Duality Discovery

A universality class groups systems by shared critical behavior. The present model has a symmetry analogy with 2D Ising. The manuscript reports β ≈ 0.22 without a bound fitting artifact, and γ remains unmeasured. These values supply no class assignment or inheritance of class-specific properties. Annex 60 discusses candidate crossover and dimensional interpretations.

The 2D Ising model possesses a deep symmetry called Kramers-Wannier duality, discovered in 1941, three years before Onsager’s exact solution. This duality maps the ordered phase to the disordered phase and vice versa, the way a photographic negative swaps dark for light while carrying exactly the same picture. The mathematics is exact:

$$\sinh(2K) \cdot \sinh(2K^*) = 1$$

where K is the coupling strength and K* is its dual. At the critical point, K = K*: the system equals its own dual.

The question for alignment: if trust is the ordered phase, what is its dual?

The Dual Order Parameter

In the original Ising picture: - Spins ($\sigma_i = \pm 1$) live on lattice sites - Magnetization $M = \langle\sigma\rangle$ measures collective alignment - Ordered phase: high $M$, spins mostly parallel

In the dual picture: - Domain walls live between sites with opposite spins - Domain-wall density counts disagreements between neighboring spins. It is distinct from the nonlocal Ising disorder operator μ, which corresponds to dual order. - Kramers-Wannier duality exchanges ordered and disordered phases at transformed couplings; a wall-density count alone does not construct that duality.

For AI alignment, this maps to: | Agent observable | Proposed physical analogy | |——————|—————————| | Collective alignment φ | Original order parameter | | Behavioral boundary density | Domain-wall density |

Behavioral boundaries are places where responses switch between aligned and misaligned states. Counting them supplies an inconsistency observable, without constructing a nonlocal disorder operator, dual partition function, or dual attractor.

What We Found

We ran numerical tests for Kramers-Wannier duality in our agent model:

(Here $\mu_c$ denotes the location of the critical point on the disorder-parameter axis, distinct from $\mu_\text{dual}$, the dual order parameter itself. $\Delta\mu_c$ is the separation between the two susceptibility peaks. These results were produced by the verification script kramers_wannier_verification.py; full numerics in KRAMERS_WANNIER_RESULTS.md.)

Verdict: A reported transition-pattern analogy; a duality mapping remains unestablished.

The elevated variance ratio (~4.4×) may reflect continuous agent states and all-to-all topology. Universality class and exact duality are separate questions. The reported β ≈ 0.22 has no bound fit here, and γ has not been independently measured. Similar symmetry and transition patterns motivate an analogy; they establish neither shared exponents nor an exact duality mapping.

The Philosophical Payoff

The reported peak in domain-wall clusters motivates monitoring boundary structure near a transition. It does not prove a dual order parameter or a Kramers-Wannier mapping. Structured misalignment can be studied through correlations and clusters without assigning it an ordered dual phase.

A genuine duality would require a specified transformation of the state variables and couplings, with the corresponding partition-function relation. That construction remains absent for the agent model.

Two Descriptions, One Physics

The deepest implication:

For the Ising system, physics provides equivalent descriptions at dual couplings, exchanging order and disorder. Applying that exchange to alignment and misalignment remains an analogy until a dual transformation for the agent model is constructed.

Neither description is more fundamental. The underlying reality is one thing; our descriptions of it are choices.

This is why ethics cannot be reduced to physics.

Physical dynamics and ethical preference answer different questions. A specified model can compare the stability of its equilibria. That comparison alone does not identify which equilibrium ought to be valued.

Kramers-Wannier duality exchanges coupling regimes. It proves neither simultaneous trust and defection attractors nor greater depth of a trust basin at the original parameters. Cartels and deterrence arrangements illustrate possible coordination under their own incentives; their existence needs a game-specific analysis.

Practical Implications

Alignment training targets aligned outputs. Consistency training could separately penalize inappropriate variation across matched inputs. A boundary-density statistic is a candidate monitor, with legitimate context sensitivity kept distinct from failure.

These objectives need direct robustness comparisons. No duality theorem here establishes that one resists noise and the other systematic attacks.

Jailbreaks as Domain Wall Injection

A jailbreak can create localized behavioral inconsistency. Treating that inconsistency as a domain boundary is an analogy, without an established dual phase.

Practical implication: monitoring domain wall density (behavioral inconsistency) may detect jailbreaks before alignment metrics collapse.

The Dual Trust Attractor

A stable Trust Attractor does not imply a second stable attractor at the same parameters. Coordinated defection requires its own game, payoff structure, or dynamical model. The examples below illustrate possible defection equilibria rather than consequences of Kramers-Wannier duality.

Examples in human systems: - Cartels (coordinated non-competition) - MAD (mutual assured destruction) - Stable adversarial equilibria in game theory

These examples have coordination structure under their own incentives. They establish neither dual order nor a universal ranking of basin depths. Their stability and ethical acceptability need separate arguments.

Summary

The Kramers-Wannier investigation revealed:

1. Exact duality was not established for the continuous all-to-all model.
2. Boundary density is a useful observable, distinct from a proved dual disorder operator.
3. Misalignment can have structure without forming an ordered dual phase.
4. Stability and ethical value require separate arguments; the duality discussion supplies no universal basin-depth ranking.

Annex 60 motivates a candidate branching ratio R_B ≈ 4 from network analogies. It supplies no computed RG transformation, fixed-point spectrum, or derivation of that value, and therefore establishes no duality constraint on the branching ratio.

A duality, where proved, relates descriptions and parameter regimes. Its mathematical correspondence leaves the ethical evaluation to an additional argument.

Physics gives us the landscape. Ethics chooses the destination.

“What if the difference between trust and distrust were merely a choice of which dual to inhabit?”

The answer requires a specified model of stability and a separate account of value. The present analogy settles neither.