Kramers–Wannier Duality Overview

• Kramers–Wannier duality is a transformation relating coupling constants in lattice models, mapping high-temperature behaviors to low-temperature descriptions in the Ising model.
• It employs Fourier transformation and homological analysis to rigorously establish duality on finite lattices and toroidal geometries.
• In quantum systems and gauge theories, the duality interchanges local order with non-invertible symmetry, offering insights into topological phases and computational complexity.

Searching arXiv for recent and foundational work on Kramers–Wannier duality to ground the article in the provided literature. Kramers–Wannier duality is a transformation relating a lattice model at one coupling or temperature to a dual model at another coupling. In its classical form it exchanges the high-temperature and low-temperature descriptions of the two-dimensional Ising model; in quantum formulations it exchanges order and disorder operators, interchanges couplings $J\leftrightarrow h$ in Ising-type chains, and, in the appropriate symmetry sector, maps short-range entangled states to long-range entangled duals (Al-Bashabsheh et al., 2016, Cheng et al., 2 Jul 2026).

1. Classical statement and self-dual point

For the square-lattice Ising model, Kramers–Wannier duality identifies a dual coupling $K^\ast$ through

$\sinh(2K)\,\sinh(2K^\ast)=1,$

equivalently matching the high-temperature loop expansion of the primal model to the low-temperature domain-wall expansion of the dual model. In the notation used for the toroidal formulation, the dual inverse temperature can be written as

$\widetilde\beta=-\frac12\log(\tanh\beta),$

and the free-energy density obeys

$$\lim_{n\to\infty}\frac1n\log Z(\beta)=\log c_\beta+\lim_{n\to\infty}\frac1n\log Z(\widetilde\beta), \qquad c_\beta=2\sinh\beta\cosh\beta,$$

so low temperature is mapped to high temperature and conversely (Mitchell et al., 2013, Al-Bashabsheh et al., 2016).

On self-dual lattices, the fixed point of the coupling map yields the standard critical-coupling candidate. For the square-lattice Ising model this gives

$\sinh(2K_c)=1, \qquad K_c=\frac12\log(1+\sqrt2).$

A standard misconception is that duality alone proves the full critical theory. The available derivations identify the self-dual point and the exact free-energy relation, but they do not by themselves determine scaling exponents or detailed critical behavior (Mitchell et al., 2013, Al-Bashabsheh et al., 2016).

2. Finite lattices, factor graphs, and homological sectors

A precise finite-size formulation arises by placing the Ising model on an $L\times L$ torus and rewriting it in terms of edge variables $x_e\in\mathbb Z_2$ with interaction

$$\kappa_\beta(0)=e^\beta,\qquad \kappa_\beta(1)=e^{-\beta}.$$

In the normal-factor-graph formulation, site variables live on equality nodes, edge variables on interaction factors, and parity nodes impose $x_e=x_j-x_i$. After Fourier transformation, the edge interaction satisfies

$K^\ast$0

so the dual model is again of Ising type but at dual inverse temperature $K^\ast$1 (Al-Bashabsheh et al., 2016).

The topological content enters through the chain complex of the lattice. The Fourier-transformed support is $K^\ast$2, the space of $K^\ast$3-cycles, while the original interaction representation is an image representation. On the torus,

$K^\ast$4

and the cycle space decomposes into four cosets of the boundary space: $K^\ast$5 This yields an exact finite-size duality relation in which the primal partition function equals a constant times the average of four dual partition functions, corresponding to the four topological sectors of the torus. In the thermodynamic limit these sector differences are subextensive, recovering the standard free-energy duality. In this formulation, Kramers–Wannier duality is the combination of a one-dimensional Fourier transform on local interactions and a homological rewriting from kernel support to image support (Al-Bashabsheh et al., 2016).

3. Quantum chains, boundary sectors, and non-invertible symmetry

In the one-dimensional quantum Ising chain, the textbook operator map introduces dual spins on bonds: $K^\ast$6 On a ring, however, the naive periodic mapping sends the global $K^\ast$7 charge $K^\ast$8 to the identity, so it cannot be implemented as an invertible unitary on the full Hilbert space. Exact realizations therefore require either restriction to the even or charge-neutral sector, or boundary dressings by the global symmetry operator. Within that sector, explicit purely unitary circuits implementing the full $K^\ast$9 and $\sinh(2K)\,\sinh(2K^\ast)=1,$0 duality maps can be constructed with $\sinh(2K)\,\sinh(2K^\ast)=1,$1 depth using nonlocal connectivity, whereas local circuits require linear depth and constant-depth purely unitary implementations are excluded (Cheng et al., 2 Jul 2026, Khan et al., 2024).

At criticality, the quantum duality is naturally expressed as a non-invertible symmetry rather than an ordinary group action. In lattice formulations of the critical transverse-field Ising chain, one finds operators whose square is not the identity but translation times projection onto a symmetry sector, for example

$\sinh(2K)\,\sinh(2K^\ast)=1,$2

or, in a basis-rotated description,

$\sinh(2K)\,\sinh(2K^\ast)=1,$3

Wave-function constructions in the anyonic fusion basis derive the same structure from a generalized translation built from $\sinh(2K)\,\sinh(2K^\ast)=1,$4-moves, and the resulting symmetry operator naturally takes matrix-product-operator form. In integrable trotterized chains the situation doubles: discretizing time produces two non-invertible duality operators $\sinh(2K)\,\sinh(2K^\ast)=1,$5 whose squares generate combined spatial and temporal translations, so the duality acts as a half-step along discrete light-cone directions rather than as an ordinary on-site symmetry (Sinha et al., 6 Nov 2025, Zhang et al., 2024).

4. Gauge-theoretic and topological reformulations

A topological reinterpretation places the Ising model on the boundary of a theory in one higher dimension. In this framework, Ising models are boundary theories for three-dimensional finite gauge theory, electromagnetic duality in the bulk induces Kramers–Wannier duality on the boundary, and the Ising order and disorder operators are the endpoints of Wilson and ’t Hooft defects, respectively. This viewpoint also explains why coupling to background bundles and defect insertions resolves the familiar mismatch between boundaries, cycles, and disorder lines in elementary Fourier-transform derivations (Freed et al., 2018).

The same perspective extends beyond two-dimensional Ising systems. In $\sinh(2K)\,\sinh(2K^\ast)=1,$6-dimensional gauge theories, there exist non-invertible topological defects whose fusion rules are higher-dimensional analogs of the Kramers–Wannier defect of the $\sinh(2K)\,\sinh(2K^\ast)=1,$7-dimensional critical Ising model. Explicit examples include $\sinh(2K)\,\sinh(2K^\ast)=1,$8 Yang–Mills at $\sinh(2K)\,\sinh(2K^\ast)=1,$9, $\widetilde\beta=-\frac12\log(\tanh\beta),$0 $\widetilde\beta=-\frac12\log(\tanh\beta),$1 super Yang–Mills, and $\widetilde\beta=-\frac12\log(\tanh\beta),$2 $\widetilde\beta=-\frac12\log(\tanh\beta),$3 super Yang–Mills at $\widetilde\beta=-\frac12\log(\tanh\beta),$4. In that setting, self-duality under gauging of discrete higher-form symmetries is the higher-dimensional analog of the Ising order–disorder duality (Kaidi et al., 2021).

5. Generalizations beyond the square-lattice Ising model

The algebraic and topological machinery extends to other statistical models, but not uniformly. For the three-dimensional cubic-lattice Ising model, Fourier transformation again produces a dual description, yet the dual model is no longer the same pairwise vertex-spin theory: it becomes a face-spin model with four-body interactions around cubes. On the open cubic lattice, $\widetilde\beta=-\frac12\log(\tanh\beta),$5, so there are no nontrivial topological sectors; on the three-torus, $\widetilde\beta=-\frac12\log(\tanh\beta),$6, so the duality involves $\widetilde\beta=-\frac12\log(\tanh\beta),$7 sectors. For the standard $\widetilde\beta=-\frac12\log(\tanh\beta),$8-state Potts model, when the interaction depends only on equality versus inequality, analogous Kramers–Wannier-type relations survive; for the vector Potts model with Lee-distance interactions, the Fourier transform does not generally return the same interaction class, except in special cases such as $\widetilde\beta=-\frac12\log(\tanh\beta),$9 or $$\lim_{n\to\infty}\frac1n\log Z(\beta)=\log c_\beta+\lim_{n\to\infty}\frac1n\log Z(\widetilde\beta), \qquad c_\beta=2\sinh\beta\cosh\beta,$$0 (Al-Bashabsheh et al., 2016).

On arbitrary planar graphs with bond-dependent couplings $$\lim_{n\to\infty}\frac1n\log Z(\beta)=\log c_\beta+\lim_{n\to\infty}\frac1n\log Z(\widetilde\beta), \qquad c_\beta=2\sinh\beta\cosh\beta,$$1, the free energy can be written as the determinant of local ordered and disordered operators defined on primal and dual vertices, making the duality explicit even in the random-bond setting. This determinant formulation gives a direct ordered/disordered decomposition of the planar Ising free energy and extends naturally to random-bond Ising systems (Song, 2023).

More algebraic generalizations replace the Ising interaction graph or symmetry itself. The Bilinear Phase Map formalism encodes generalized Kramers–Wannier-like transformations by a $$\lim_{n\to\infty}\frac1n\log Z(\beta)=\log c_\beta+\lim_{n\to\infty}\frac1n\log Z(\widetilde\beta), \qquad c_\beta=2\sinh\beta\cosh\beta,$$2-valued matrix $$\lim_{n\to\infty}\frac1n\log Z(\beta)=\log c_\beta+\lim_{n\to\infty}\frac1n\log Z(\widetilde\beta), \qquad c_\beta=2\sinh\beta\cosh\beta,$$3; $$\lim_{n\to\infty}\frac1n\log Z(\beta)=\log c_\beta+\lim_{n\to\infty}\frac1n\log Z(\widetilde\beta), \qquad c_\beta=2\sinh\beta\cosh\beta,$$4 determines the underlying global symmetry, rank deficiency explains the loss of unitarity, and the general fusion rule takes the form

$$\lim_{n\to\infty}\frac1n\log Z(\beta)=\log c_\beta+\lim_{n\to\infty}\frac1n\log Z(\widetilde\beta), \qquad c_\beta=2\sinh\beta\cosh\beta,$$5

In this way, generalized dualities connect spontaneous-symmetry-breaking, trivial, and symmetry-protected-topological phases into explicit duality webs (Yan et al., 2024). A different extension replaces the spin algebra by a finite-dimensional semisimple Hopf algebra $$\lim_{n\to\infty}\frac1n\log Z(\beta)=\log c_\beta+\lim_{n\to\infty}\frac1n\log Z(\widetilde\beta), \qquad c_\beta=2\sinh\beta\cosh\beta,$$6. When $$\lim_{n\to\infty}\frac1n\log Z(\beta)=\log c_\beta+\lim_{n\to\infty}\frac1n\log Z(\widetilde\beta), \qquad c_\beta=2\sinh\beta\cosh\beta,$$7 is self-dual, generalized Hopf-Ising chains admit a Kramers–Wannier operator exchanging paramagnetic and ferromagnetic phases, and the infrared symmetry is a weakly integral fusion category given by a $$\lim_{n\to\infty}\frac1n\log Z(\beta)=\log c_\beta+\lim_{n\to\infty}\frac1n\log Z(\widetilde\beta), \qquad c_\beta=2\sinh\beta\cosh\beta,$$8 extension of $$\lim_{n\to\infty}\frac1n\log Z(\beta)=\log c_\beta+\lim_{n\to\infty}\frac1n\log Z(\widetilde\beta), \qquad c_\beta=2\sinh\beta\cosh\beta,$$9. For the Kac–Paljutkin algebra $\sinh(2K_c)=1, \qquad K_c=\frac12\log(1+\sqrt2).$0, numerical analysis identifies four of the six $\sinh(2K_c)=1, \qquad K_c=\frac12\log(1+\sqrt2).$1-symmetric gapped phases, separated by Ising critical lines and meeting at a multicritical point, while lattice comodule-algebra constructions realize all six phases (Lu et al., 10 Feb 2026).

6. Computational realizations and contemporary applications

Kramers–Wannier duality has also been recast in explicitly operational language. On the transverse-field Ising ring, the duality can be implemented as a superoperator acting on the tensor product of the Ising and dual-Ising Hilbert spaces, and the relation between this superoperator and the Hilbert-space duality map is naturally expressed as a quantum operation with an operator-sum representation. In that framework, the well-known fusion rules are reproduced directly at the level of quantum channels rather than only at the level of formal defect algebra (Khan et al., 2024).

Experimental realizations have begun to target the defect structure itself. A duality-twisted Floquet Ising chain binds an isolated zero mode at the duality defect; the mode is localized at the domain wall between regions related by a Kramers–Wannier transformation, has an infinite lifetime in the exactly solvable setting, and was generated on a digital quantum computer using Floquet driving of a closed Ising chain with a duality defect. Its associated persistent autocorrelation provides a direct dynamical signature, and the mode remains robust against integrability- and symmetry-breaking perturbations in finite systems (Samanta et al., 2023).

The duality has also migrated into domains far from its original thermodynamic setting. In computer science, a Kramers–Wannier-like duality for Ising systems with multi-spin interactions maps $\sinh(2K_c)=1, \qquad K_c=\frac12\log(1+\sqrt2).$2SAT to counting non-negative solutions of a linear Diophantine system, making the complexity of the dual counting problem explicit (Mitchell et al., 2013). In many-body dynamics, sequential quantum circuits implementing Kramers–Wannier duality have been used to analyze quantum many-body scar states: when the duality preserves the embedding conditions for scarring, scar states are mapped to dual scar states, whereas in twisted sectors lacking those conditions the dual images are thermal states. In that setting, non-invertible duality functions as a diagnostic of scar stability rather than only as an order–disorder map (Fontana et al., 7 Aug 2025).

Across these formulations, the persistent invariant is structural rather than merely computational. Kramers–Wannier duality exchanges image and kernel descriptions on lattices, order and disorder operators in quantum chains, Wilson and ’t Hooft defects in topological field theory, and symmetry data before and after gauging. Its modern generalizations retain the same formal pattern while broadening the ambient setting from square-lattice Ising models to planar disorder, higher-dimensional gauge theory, non-invertible categorical symmetry, quantum circuits, and algorithmic counting problems.