Machine Learning for Model Universality Class Recognition

We study the critical behavior of models in statistical mechanics: 4-state Potts, Baxter-Wu, Ashkin-Teller, and Turban. Models have different lattice geometry and Hamiltonians, although it belongs to the same universality class. The thermodynamic functions behave similarly near the critical point, and functions singularity governs by the same critical exponents. The paper [10] proposes using machine learning to extract the universal behavior of the triangular Ising model training the network with the square Ising model. The models differ only by the lattice geometry and have the same Hamiltonian. We extend their approach to the case of the four different models in the same universality class. We extend the method [10] to estimate the accuracy of the approach. We present preliminary results applying machine learning to the four above mentioned models.