Purity of thermal mixed quantum states

Abstract

We develop a formula to evaluate the purity of a series of thermal equilibrium states that can be calculated in numerical experiments without knowing the exact form of the quantum state \textit{a priori}. Canonical typicality guarantees that there are numerous microscopically different expressions of such states, which we call thermal mixed quantum (TMQ) states. Suppose that we construct a TMQ state by a mixture of $N_\mathrm{samp}$ independent pure states. The weight of each pure state is given by its norm, and the partition function is given by the average of the norms. To qualify how efficiently the mixture is done, we introduce a quantum statistical quantity called "normalized fluctuation of partition function (NFPF)". For smaller NFPF, the TMQ state is closer to the equally weighted mixture of pure states, which means higher efficiency, requiring a smaller $N_\mathrm{samp}$. The largest NFPF is realized in the Gibbs state with purity-0 and exponentially large $N_\mathrm{samp}$, while the smallest NFPF is given for thermal pure quantum state with purity-1 and $N_\mathrm{samp}=1$. The purity is formulated using solely the NFPF and roughly gives $N_\mathrm{samp}^{-1}$. Our analytical results are numerically tested and confirmed by the two random sampling methods built on matrix-product-state-based wave functions.