Berezinskii-Kosterlitz-Thouless phase transitions of the antiferromagnetic Ising model with ferromagnetic next-nearest-neighbor interactions on the kagome lattice

We investigate the six-state clock universality of the Ising model on the kagome lattice, considering antiferromagnetic nearest-neighbor (NN) and ferromagnetic next-nearest-neighbor (NNN) interactions. Our comprehensive study employs three approaches: the level-spectroscopy method, Monte Carlo simulations, and a machine-learning phase classification technique. In this system, we observe two Berezinskii-Kosterlitz-Thouless (BKT) transitions. We present a phase diagram consisting of three phases: the low-temperature ordered phase with sublattice magnetizations, the intermediate BKT phase, and the high-temperature disordered phase, as a function of the ratio of the NNN interaction to the NN interaction. We verify the six-state clock universality through the machine-learning study, which uses data from the six-state clock model on the kagome lattice for training.

💡 Research Summary

In this work the authors investigate a two‑dimensional Ising model defined on the kagome lattice with antiferromagnetic nearest‑neighbor (NN) coupling J₁ ≥ 0 and ferromagnetic next‑nearest‑neighbor (NNN) coupling J₂ ≥ 0. While the pure NN antiferromagnet (J₂ = 0) is exactly solvable and possesses a highly degenerate, disordered ground state with a large residual entropy, the addition of a ferromagnetic NNN term lifts this degeneracy and stabilizes a six‑fold ordered phase. The ordered phase can be described by a complex order parameter m = |m| e^{iπσ/2} e^{i2π α/3} (σ = 0,1; α = 0,1,2), which exhibits a Z₆ symmetry identical to that of the six‑state clock model. Consequently, the system is expected to belong to the six‑state clock universality class and to display two Berezinskii‑Kosterlitz‑Thouless (BKT) transitions: a low‑temperature BKT transition T₁ at which the Z₆ symmetry is spontaneously broken, and a higher‑temperature BKT transition T₂ at which vortex–antivortex pairs unbind, leading to a disordered phase.

To verify this scenario the authors employ three independent numerical approaches:

1. Level‑spectroscopy (LS) of transfer matrices – The row‑to‑row transfer matrix T is constructed for cylindrical systems with periodic boundary conditions. By exploiting translation by two lattice spacings, spin‑reversal, and inversion symmetries, the spectrum is decomposed into sectors labeled by momentum k, parity T, and inversion P. From the leading eigenvalues λₙ(k,T,P) the finite‑size scaling dimensions x₀(l) and x_S(l) are extracted (l = ln L). The low‑temperature BKT point is located by the level‑crossing condition x₀(l) = 4 x_S(l), while the high‑temperature BKT point satisfies \bar{x}_0(l) = (16/9) x_S(l). Finite‑size estimates T₁(L) and T₂(L) are extrapolated to the thermodynamic limit using the known O(L⁻²) correction, which is free of logarithmic terms because the marginal operator is absent at the crossing.

2. Monte‑Carlo (MC) simulations with parallel‑tempering – Large‑scale simulations are performed for L = 24, 36, 48 (and larger) using replica‑exchange to overcome critical slowing down. The spin‑spin correlation function g(r) = ⟨s_i s_{i+r}⟩ is measured, and the ratio R(T) = g(L/2)/g(L/4) is formed. In a BKT phase the correlation length ξ diverges, making R(T) size‑independent; deviations from size‑independence signal crossing of T₁ or T₂. By collapsing data for different L the authors obtain precise estimates of the two transition temperatures.

3. Machine‑learning (ML) phase classification – A fully connected neural network (100 hidden units) is trained on spin configurations generated from the pure six‑state clock model on the kagome lattice. Training data are taken from the three known phases (ordered, BKT, disordered). After training, the network is fed configurations of the antiferromagnetic kagome Ising model with NNN coupling across a range of temperatures. The network outputs probabilities for each phase; the temperatures where the probabilities cross 0.5 provide independent estimates of T₁ and T₂. Cross‑entropy loss with L₂ regularization and the Adam optimizer ensure robust training without over‑fitting.

All three methods give mutually consistent transition temperatures. For a representative coupling ratio J₂/J₁ = 1/3 the authors find T₁ ≈ 0.12 J₁/k_B and T₂ ≈ 0.28 J₁/k_B (values vary smoothly with J₂/J₁). The phase diagram plotted in the (u = e^{−J₁/T}, v = e^{J₂/T}) plane shows three regions: a low‑temperature ordered phase with sublattice magnetizations m_α, an intermediate critical BKT phase with algebraic correlations, and a high‑temperature disordered phase. The ordered phase exhibits a nine‑site magnetic unit cell and breaks both time‑reversal and sublattice symmetries, consistent with the Z₆ order parameter.

The authors interpret the results within the dual sine‑Gordon field theory. The operator √2 cos 6√2 φ (locking the phase field φ) becomes relevant at T₁, generating the six‑fold ordered state, while the dual operator √2 cos √2 θ (locking the dual field θ) becomes relevant at T₂, driving the vortex unbinding transition. The agreement between the numerical data and the predictions of the six‑state clock universality confirms that the kagome antiferromagnetic Ising model with ferromagnetic NNN interactions provides a concrete lattice realization of a BKT transition emerging from a frustrated, disordered ground state.

In summary, this study demonstrates that adding a ferromagnetic next‑nearest‑neighbor coupling to the kagome antiferromagnetic Ising model lifts the macroscopic degeneracy, stabilizes a six‑fold ordered phase, and produces two BKT transitions separating ordered, critical, and disordered regimes. The work showcases the power of combining analytical field‑theory insight, exact transfer‑matrix level spectroscopy, large‑scale Monte‑Carlo, and modern machine‑learning techniques to resolve subtle topological transitions in highly frustrated systems. The findings broaden the class of lattice models known to exhibit BKT physics and suggest routes for engineering similar phenomena in experimental magnetic materials where competing interactions can be tuned.