Benchmarking the simplest slave-particle theory with Hubbard dimer

Abstract

Slave-particle method is a powerful tool to tackle the correlation effect in quantum many-body physics. Although it has been successfully used to comprehend various intriguing problems, such as Mott metal-insulator transition and Kondo effect, there is still no convincing theory so far on the availability and limitation of this method. The abuse of slave-particle method may lead to wrong physics. As the simplest slave-particle method, $\mathbb{Z}_2$ slave spin, which is widely applied to many strongly correlated problems, is highly accessible and researchable. In this work, we will uncover the nature of $\mathbb{Z}_2$ slave-spin method by studying a two-site Hubbard model. After exploring some properties of this toy model, we make a comparative analysis of the results obtained by three methods: (i) slave-spin method on mean-field level, (ii) slave-spin method with gauge constraint and (iii) the exact solution as a benchmark. We find that, protected by particle-hole symmetry, the slave-spin mean-field method can recover the static properties of ground state exactly at half filling. Furthermore, in the parameter space where both $U$ and $T$ are small enough, slave-spin mean-field method is also reliable in calculating dynamic and thermal dynamic properties. However, when $U$ or $T$ is considerably large, the mean-field approximation gives ill-defined behavior, which results from the unphysical states in enlarged Hilbert space. These findings lead to our conclusion that the accuracy of slave particle can be guaranteed if we can exclude all unphysical states by enforcing gauge constraints.Our work demonstrates the promising prospect of slave-particle method in studying complex strongly correlated models with specific symmetry or in certain parameter space.